Economic Factors' Impact on Construction Cost Estimation: A Deep Neural Network Study, Schemes and Mind Maps of Construction management

Download Economic Factors' Impact on Construction Cost Estimation: A Deep Neural Network Study and more Schemes and Mind Maps Construction management in PDF only on Docsity! Investigation of Factors Impacting Construction Cost Estimate to Develop Construction- Driven Artificial Neural Network (ANN) by Salem Al Saber A Thesis Presented in Partial Fulfillment of the Requirements for the Degree Master of Science Approved July 2023 by the Graduate Supervisory Committee: Kenneth Sullivan, Chair Richard Standage Kristen Hurtado ARIZONA STATE UNIVERSITY August 2023 i ABSTRACT Construction industry is the backbone of any country’s economy. It is a primary source of foreign investments, creates new jobs, and maintains the economy flowing in various trades. Accurate cost estimation is a critical aspect for the construction industry, directly impacting project success and profitability. This master's thesis focuses on comprehensively identifying the key factors that influence cost estimation and provides valuable recommendations for constructing an optimized Artificial Neural Network (ANN) model. Through an extensive research methodology encompassing literature review, surveys, and interviews with industry professionals, this study uncovers significant factors that exert a substantial impact on cost estimation practices. The findings emphasize the importance of seamlessly integrating project delivery systems, meticulously considering project duration, and incorporating diverse perspectives from global regions. By incorporating these insights, stakeholders can make informed decisions, enhance project planning, and elevate overall project performance. This study successfully bridges the gap between theory and practice, presenting invaluable insights for stakeholders within the construction industry. Keywords: cost estimation, construction industry, Artificial Neural Network, factors, project delivery systems, project duration, global perspectives, informed decision-making, project planning, project performance iv TABLE OF CONTENTS Page LIST OF TABLES .................................................................................................................... v LIST OF FIGURES ................................................................................................................. vi LIST OF ABBREVIATIONS .................................................................................................. x CHAPTER 1. INTRODUCTION ........................................................................................................ 1 2. LITERATURE REVIEW .......................................................................................... 10 3. RESEARCH METHODOLOGY .............................................................................. 60 4. DATA COLLECTION AND RESULTS .................................................................. 67 5. CONCLUSION AND RECOMMENDATIONS ................................................... 101 REFERENCES .................................................................................................................... 105 APPENDIX A SURVEY SAMPLE USED TO COLLECT STUDY RESULTS ........................ 112 B IRB FORM ............................................................................................................. 129 v LIST OF TABLES Table Page Table 1: Key Influential Factors Examined in Previous Research ................................... 65 Table 2: Top 10 Factors for Impact on Cost Estimation. .................................................. 95 vi LIST OF FIGURES Figure Page Figure 1: Project Costs Classification ............................................................................... 17 Figure 2: The Structure of a Construction Company’s Overhead Costs. Extracted From (Apanavičienė & Daugėlienė, 2011)................................................................................. 20 Figure 3: Schematic Diagram of a Neuron. Extracted From (Najafi & Tiong, 2015) ...... 33 Figure 4: The Structure of Artificial Neural Network. Extracted From (Jinisha & Jothi, 2019) ................................................................................................................................. 34 Figure 5: Diagram of Learning With a Teacher. Extracted From (Haykin, 2009) ........... 36 Figure 6: Diagram of Reinforcement Learning. Extracted From (Haykin, 2009) ............ 37 Figure 7: The Diagram of Artificial Neural Network Function: Threshold. Extracted From (Artificial Neural Network, n.d.) ...................................................................................... 38 Figure 8: The Diagram of Artificial Neural Network Function: Type Sigmoid. Extracted From (Artificial Neural Network, n.d.) ............................................................................. 39 Figure 9: The Diagram of Artificial Neural Network Function: Piecewise Linear. Extracted From (Artificial Neural Network, n.d.) ............................................................................. 39 Figure 10: The Diagram of Artificial Neural Network Function: Gaussian. Extracted From (Artificial Neural Network, n.d.) ...................................................................................... 40 Figure 11: Single Layer Feed Forward Network. Extracted From (Chakraverty et al., 2019) ........................................................................................................................................... 42 Figure 12: Single layer feed forward network. Extracted from (Najafi & Tiong, 2015) .. 43 Figure 13: Multilayer Perceptron (MLP) Showing Backpropagation Algorithm. Extracted From (Alcineide et al., 2021) ............................................................................................ 44 ix Figure Page Figure 38: Population Rating Impact of Weather on Cost Estimation (Total Sample of 87). ........................................................................................................................................... 87 Figure 39: Population Rating Impact of Management Factor on Cost Estimation (Total Sample of 87). ................................................................................................................... 88 Figure 40: Population Rating Impact of Availability of Labor Factor on Cost Estimation (Total Sample of 87). ........................................................................................................ 89 Figure 41: Population Rating Impact of Type of Client Factor on Cost Estimation (Total Sample of 87). ................................................................................................................... 90 Figure 42: Population Rating Impact of Market Trend Factor on Cost Estimation (Total Sample of 87). ................................................................................................................... 90 Figure 43: Population Rating Impact of External Social Conditions Factor on Cost Estimation (Total Sample of 87). ...................................................................................... 91 Figure 44: Population Rating Impact of Project Sustainability Level on Cost Estimation (Total Sample of 87). ........................................................................................................ 92 Figure 45: Population Rating Impact of Safety Requirements Factor on Cost Estimation (Total Sample of 87). ........................................................................................................ 93 Figure 46: Population Rating Impact of Labor Union Availability on Cost Estimation (Total Sample of 87). ........................................................................................................ 94 x LIST OF ABBREVIATION ANN - Artificial Neural Network BIM - Building Information Modeling UAE - United Arab Emirates MLP - Multi-Layer Perceptron RNN - Recurrent Neural Network PMI - Project Management Institute AACE - Association for the Advancement of Cost Engineering EVM - Earned Value Management CBR – Case Base Resonance MAE – Mean Absolute Error MSE – Mean Squared Error SGD – Stochastic Gradient Descent MLP- Multilayer Perceptron 1 CHAPTER 1 INTRODUCTION Background A construction project can be defined successful by being limited within the triple constraint or the golden triangle, which encompasses the interrelated factors of budget, time, and quality. Accurate cost estimation plays a crucial role in enabling stakeholders and decision-makers to conduct comprehensive feasibility studies, determine appropriate financial scales during the bidding stage, and effectively monitor cash flows throughout the construction phase (Elmousalami, 2020; Matel et al., 2019; Wang et al., 2022). It is imperative to minimize errors in construction cost estimation as accurate information empowers stakeholders in construction engineering and management to make well-informed decisions aligned with the project's financial requirements (Wang et al., 2022). Moreover, accurate cost estimation plays a pivotal role in determining whether to proceed or to cancel the project (Matel et al., 2019). Each construction project possesses unique characteristics, and the estimation of resource utilization and specification choices significantly impacts its associated costs. Throughout the various stages of a building project's life cycle, designing teams encounter risk and uncertainty. Therefore, it is essential to duly recognize and account for all relevant factors in order to mitigate the potential for project delays and cost overruns (Alqahtani & Whyte, 2016). 4 affecting cost estimation are the backbone of any ANN model, as they characterize the input variable that shapes the output variable. The chosen factors should have the biggest impact in affecting cost estimations, where the ANN model will get trained on these factors and will reduce their uncertainty. The current body of knowledge focus on basic factors such as project locations, various elements of the structure, with minor considerations to the construction index, usage of advanced technology, linking usage of Building Information Modeling (BIM) and advanced technologies to mitigate delays and abortive works, social environment, and international market conditions. These neglected factors could have higher impact on cost estimation which is why ANN should not be trained only on structural elements factors such as number of stories, type of foundation, etc. Research Aim & Objectives The aim of this thesis is to identify and prioritize key factors that have a significant impact on cost estimation for building projects in the United Arab Emirates (UAE). These factors will have value as they represent the input variables for the next stage where an ANN model will be developed with the aim of providing an efficient and accurate cost estimation method. The research will involve an extensive analysis of existing literature, as well as conducting a survey and interviews to develop the survey results. By identifying the most influential and relevant factors, this thesis seeks to enhance the accuracy, efficiency, and interpretability of ANN models specifically tailored for the construction industry. The findings of this research will benefit various stakeholders involved in construction projects, such as owners, contractors, and consultants, by enabling them to 5 determine parameters that affect cost estimation and can cause project cost overruns. The ultimate goal of this study is to prepare input parameters that will enable to provide valuable insights for future development of ANN models in the field of construction. The Main objectives of this thesis can be defined under the two points below: 1. Identify new key factors affecting the accuracy of estimation of building project within the UAE and similar markets with focus on structural factors and new prospective factors such as safety requirement, type of contract, usage of advanced techniques, etc. 2. Conduct a comprehensive analysis for the determined key factors, followed by usage of tools for the development of data collection using various methods such as pilot testing and expert reviews that will be further discussed in chapter three. Expert reviews are conducted with high management professionals such as planning manager, contracts manager, technical manager, estimation manager and general manager to prepare for the development of the optimum Neural Network model in the next phase. Significance of the Study The thesis analyzes cost factors by weighting and rating of the major structural works in building projects and other economic factors such as high safety requirement, cost index, usage of BIM, stakeholder characteristics, etc. The anticipated contributions of this thesis are expected to have relevance to both researchers and practitioners in the following ways: 6 1. For researchers, the findings will serve to emphasize the importance of exploring new parameters in cost estimation in order to assess the accuracy of Artificial Neural Network models across of building projects. By identifying and investigating these parameters, this research will contribute to expanding the knowledge and understanding of ANN model performance in the construction domain. 2. For practitioners, the findings will provide valuable insights to enhance the accuracy of cost estimation practices. By highlighting overlooked parameters that can lead to underestimation or overestimation of project costs, practitioners will be better equipped to conduct more precise estimate jobs. For instance, the use of BIM can offer a clearer deconstruction of the project compared to traditional 2D drawings, thereby reducing errors. However, it should be noted that the incorporation of BIM may come with additional costs related to the elaboration of the BIM model. It is the case that many estimators fail to precisely estimate the requirement to produce such models resulting in unpredicted costs, delays because of necessary qualified professionals and more serious implications. Overall, this research aims to bridge the gap between research and practice by offering practical implications for practitioners while contributing to the advancement of knowledge in the field for researchers. Research Scope and Limitation This research primarily concentrates on gathering surveys from professionals who possess significant experience in building projects within the UAE. The scope of the study 9 potential challenges associated with ANNs. The chapter concludes by depicting applications review of ANN in the construction field in recent years. Chapter 3: Research Methodology The third chapter describes the methodology adopted for developing the thesis, including the process of acquiring data on key factors affecting cost estimating in building projects. It then briefs the tools for data development. Chapter 4: Data Results In this chapter, survey results statistical analysis is presented. Then, tools for data development used during study are discussed to confirm the data reliability. Chapter 5: Conclusion and Recommendations The final chapter summarizes the conclusions drawn from the study and provides recommendations for the next phase in which an ANN model will be build and cost estimation model at early stage will be the end result. It also outlines areas for further research and development. 10 CHAPTER 2 LITERATURE REVIEW Chapter Introduction The objective of this chapter is to provide a review of the literature used in the thesis. The chapter communicates cost-related definitions, purpose, accuracy, and types of cost estimation. It also depicts the estimating process and its methods and describes the classification of construction costs. The chapter concludes by providing Artificial Neural Networks (ANNs) history, definition, structure, types, problems and challenges, and applications of ANN. Cost Engineering Cost engineering can be defined as applying scientific principles and methods along with engineering expertise and judgment, cost engineering to address a variety of issues relating to estimation, cost control, business planning, and profitability analysis (Jinisha & Jothi, 2019). Cost Estimate A cost estimate may be defined as the approximation of an operation, project, or program cost. The cost estimate is determined from a cost estimating process. The cost estimate may represent a single total value or may have many identifiable component values. 11 According to the Association for the Advancement of Cost Engineering International (AACE), Cost estimate provides the basis for project management, business planning, budget preparation, and cost and schedule control (AACE, 2020). Purpose and Accuracy of Cost Estimate Cost estimation accuracy is an essential factor for ensuring the success of any construction project. One of the major problems is cost overruns, especially with the current priority on tight budgets. A number of literature estimate that one in four projects can face overbudget issues. Many of the shelving or cancellation of projects are led by cost overruns (Feng & Li, 2014). Hanna et al. (2004) indicated that the recommendation to eliminate the top three most common reasons for change orders, design errors, design changes, and additions, is to invest additional time and budget resources ahead of construction (Hanna et al., 2004). Construction early-stage cost estimation, along with its accuracy, is one of the major challenges for decades. Initially, limited information is available, the estimation process becomes a very hard task for estimators and project engineers. An appropriate estimation can provide a clear vision for financial management in designing the budget for investors, hence making better decisions. It also provides project managers with the ability to manage their available resources and cash reserve funds during the entirety of project execution stages (Chandanshive & Kambekar, 2019). According to Kim et al. (2004), one of the main elements of decision making at the preliminary stage of construction is cost estimate. Hence, improved techniques will provide improved control of cost and time. 14 According to Amade et al. (2015), the probability of the use of each of the estimate methods depends on the ease of its application, familiarity, and effectiveness along with a tolerable level of accuracy and reliability. They include, but are not limited to: • Functional Unit: Also called unit-price. This method involves the usage of a single functional unit multiplied by the number of units used. • Cube Method: Multiplying plinth area with the height of the building. The height of the building should be considered from floor level to the top of the roof level. It is more suitable for multi-storied buildings. • Superficial Area: It is one of the most popular preliminary estimating methods. It is an approximate cost obtained by multiplying the area by the cost per square meter/feet. • Superficial Parameter: It entails the analysis of cost, programmatic and technical data to identify cost drivers and develop cost models. • Approximate Quantities: It is an approximate quantity method cost estimate, the total wall length of the structure is measured, and this length is multiplied by the rate per running meter which gives the cost of the building. • Elemental Analysis: Since the 1950s, quantity surveyors have used this technique to base their predictions during the design stage. Total costs should be provided for each element and sub-elements as appropriate. Costs should be shown separately where required in the elemental definitions and for different forms of construction. The cost of the elements should total to the contract sum minus main contractor’s profit, where identified; preliminaries; contingencies and, where appropriate, 15 contractor’s design fees. The cost of each element and the items comprising it should correspond with the specification (BCIS, 2012). Estimate methods are classified differently by various literature. Classified estimate method types are (Samphaongoen, 2010): 1. Conceptual cost estimate during the schematic design phase and the error percentage reaches to 20% 2. Semi-detailed cost estimate during the detailed design phase and the error percentage is between 5-10%. 3. Detailed cost estimate during the issue for construction phase and the error percentage decreases to less than 5% The first estimate type guideline was developed in 1958 by AACE Estimating Methods Committee and proposed four types: 1. Order of Magnitude estimate 2. Preliminary estimate 3. Definitive estimate 4. Detailed estimate All these types are based on four estimate characteristics: purpose, accuracy, information available for estimating, and methods. The main tenet in estimation is a level of scope definition increased which improves the accuracy of the estimate. While the specifics have evolved, the general concept of classification or phased estimates remains the same. 16 Estimating Process Estimation is one of the essential steps in the project management process. The process of estimating refers to the procedure of approximating or calculating the costs, resources, time, and other factors associated with a particular project (Elfaki et al., 2014). As per Project Management Institute (PMI), the estimating process involves "developing an approximation or estimate of the costs of the resources needed to complete project activities" (Project Management Institute, 2017, p. 246). The Project Management Body of Knowledge (PMBOK) also emphasizes the importance of updating estimates throughout the project life cycle to reflect changes in project scope, risk, and other factors. Elfaki et al. (2014) states that the process of estimating costs is of utmost importance in the initial stages of any construction project. As a result, the field of construction management has invested significant research efforts in the realm of construction cost estimation. Classification of Construction Costs RAD (2002) states that construction costs are generally classified into six categories: direct and indirect costs, overhead, contingency, allowance, and profit. Kraus (2007) describes tools for mutual understanding classification of construction costs, specifically overhead, contingency and allowance. It is generally agreed that contracting is a job for the lowest offer, that is why it is important that contractors should understand calculating the indirect cost, profit, and overhead. Figure 1 below simplifies the classification. 19 (Scaffolding), sick leave, vacation, training, and even retirement benefits for the employees. (RAD, 2002) Kraus (2007) states that indirect cost can be considered as site overhead and is defined as per AACE International as: All non-direct costs necessary to properly perform the installation and may include field administration, direct supervision, capital tools, startup costs, contractor’s fees, insurance, taxes, etc. Overhead Overhead costs are the costs of running a business, they are not specific and vary from one company to other for what is considered. RAD (2002) states that overhead costs can include compensation of the company senior management and the cost of infrastructure necessary for supporting project activities. He also debates items such as the cost of preparing unsuccessful proposals, marketing and public relations, and ongoing innovative ventures of the organization are included. Kraus (2007) defined overhead as per AACE International’s Recommended Practice No. 10S-90, A cost related to performing a task or a project that cannot be assigned or attributed to any part of the project. He classifies overheads into 2 sub-categories, general overhead, and site overhead. General overhead could be costs such as office, plant, equipment, staffing, and expenses that are essential to be maintained by a contractor for general business operations. According to Kraus (2007), the difference between general overhead and site overhead is often misunderstood. Site overhead are indirect cost and is defined in above section. 20 Apanavičienė & Daugėlienė (2011) stated that contractors often fail to evaluate properly the actual overhead costs, which represent the largest part of indirect costs of construction. The authors provided figure 2 below to simplify the matter. Figure 2: The Structure of a Construction Company’s Overhead Costs. Extracted From (Apanavičienė & Daugėlienė, 2011) Contingency Kraus (2007) defined Contingency as per AACE International as determined amount or percentage added to estimates to account for items, conditions, or events whose condition, occurrence, and/or impact are uncertain and likely to result in overall additional costs. The percentage depends generally on previous experience and volume of inaccuracies in the estimate caused by uncertainties in project details (RAD, 2002). It is generally included in most estimates and is expected to be expanded by the direction of the management (Kraus, 2007). 21 Allowance RAD (2002) states that allowance and contingency are akin to each other, only that allowance is defined as lump sum amount reserved for certain project task that are not calculated in a detailed estimate but is known to occur through the project. Contingency is an amount set for inaccuracy in estimate or change in price from tender stage to the procurement stage. Kraus (2007) further explains allowance as per AACE International as “additional resources included in estimates to cover the cost of known but undefined requirements for an individual activity, work item, account, or subaccount.” Profit Kraus (2007) explains profit as the amount included in a contractor’s tender price to compensate the contractor for undertaking the risks associated with the project and to provide a return on its investment. He further states that overhead and profit are frequently combined in a contract sense and referred to as “contractor’s overhead and profit”. Shehatto (2013) defines profit as a percentage of the total contract price, or in some cases, as a percentage of each task in the project. The percentage of profit is set by the higher management or owners for each individual tender, depending on local market conditions, competition, risk, and the contractor`s need for new projects. Cost Estimation Techniques Numerous literatures provide several techniques of cost estimates that are used in different industries and fields. Shehatto (2013) stated three main categories for cost estimation techniques: 24 often used when there is limited information available about the new project, and the historical data of similar projects used as a basis for estimating the cost of the new project (Mendes et al., 2003; Khosrowshahi, 1994). Various literatures proposed different variations of analogy-based techniques, including the following: 1. Analogy-based parametric models (ABPMs). These models rely on identifying the key cost drivers of the project and using them to develop a mathematical formula that can be applied to estimate the cost of similar projects. (Ruwanpura & Bandara, 2009) 2. Analogy-based cost estimation (ABCE): ABCE is a technique that involves using a database of historical cost data to identify similar projects and estimate the cost of new projects based on their similarity. This technique involves a manual search process to identify relevant analogies that may require some level of expert judgment (Nguyen, Deeds, & Carpenter, 2014). 3. Analogy-based effort estimation (ABEE): ABEE is a technique used to estimate the effort required to complete a software development project based on the similarity of the project to past projects. This technique involves identifying and measuring the characteristics of past projects and using them to estimate the effort required for the new project. (Mendes et al., 2003) Delphi Method. is described by Elmousalami (2020) as a structured communication technique used to gather the opinions of experts regarding a specific case, with the aim of identifying all the parameters that impact the 25 system. It involves a series of iterative rounds of collecting, ranking, and revising the parameters based on the feedback received from the experts. Experts provide anonymous feedback and revise their opinions based on the feedback from other participants to improve the quality of the survey. The Delphi rounds continue until a consensus is reached and no other opinions remain (Elmousalami, 2020). Davé (2003) adds that "Traditional Delphi Method (TDM) is an anonymous, written, iterative survey method used to develop consensus among experts regarding a specific topic" (Davé, 2003, p. 135). Preliminary and Detailed Techniques A preliminary estimate is an approximation that relies on specific cost data and predefined guidelines. It allows the owner to review the design before proceeding with further details. The accuracy of this estimate is typically within a range of -30% to +50% (Jinisha & Jothi, 2019). Detailed estimation techniques involve a series of methods and procedures that are employed to obtain an accurate cost estimate of a project based on analysis of detailed design specifications and drawings. These techniques entail deconstructing the project into its constituent elements, quantifying the required resources, considering factors such as labor, materials, equipment, and overhead costs, and incorporating current market rates and other pertinent variables to ensure accuracy in cost estimation (Jha & Bhandari, 2016). 26 Traditional and Artificial Intelligence Based Techniques The traditional percentage model is arbitrary and difficult to justify (Kwon & Kang, 2019). Günaydın & Doğan (2004) stated many types of traditional methods such as traditional detailed breakdown cost estimation; simplified breakdown cost estimation; cost estimation based on cost functions; activity-based cost estimation; cost index method; and expert systems. Other types can be found in different literatures such as analogue estimating (CII, 1996), and Three-Point Estimating (Fleming & Koppelman, 2016). Other methods are Monte Carlo simulations, and a third method is the regression model. These are advanced statistical tools using analytical predictions in forecasting the final cost of a project. These methods lack consideration of estimating risk costs (Kwon & Kang, 2019). Alternative approaches were introduced recently using the concept of parametric models based on computerized techniques, such as fuzzy expert systems (also called fuzzy logic) and Artificial Neural Networks (ANN), which have been utilized in creating a model for estimating project cost contingency. These models are well-suited for handling non- linear data modeling, in contrast to linear methods like regression. While they prove valuable in assessing risks and their associated probabilities, they exhibit limited effectiveness when it comes to estimating cost contingency. (Kwon & Kang, 2019) Artificial Neural Networks (ANN) Shehatto (2013) summarizes that Artificial Neural Networks (ANN) draw inspiration from the structure and functioning of neurons in the human brain. The brain demonstrates remarkable abilities to perform complex tasks with relative ease compared to computers. Consequently, researchers sought methods to integrate this intelligence into 29 Advantages of ANN Artificial Neural Networks are among the innovative techniques that excel in handling incomplete data sets, fuzzy or incomplete information, and highly complex and ill-defined problems. ANNs have the capability to learn from examples and effectively address non-linear problems. A notable feature of ANN is their capacity to learn from experience and adapt to dynamic situations. They possess a natural ability to store experiential knowledge and make it readily accessible for application (Shehatto, 2013). Matel et al. (2019) stated that in the early design stage of construction projects, ANNs are widely used to estimate project costs and duration. ANN are beneficial because they can self-learn, which saves development time, and they can identify non-linear relationships between cost factors and project cost without extra effort. By using an ANN model, accurate predictions can be obtained, even when there is limited information available during the early stages of the design process. Wang et al. (2022) noted that the regression analysis method is characterized by its simplicity and the generation of straightforward predictions. However, it has limitation that its reliance on a predetermined mathematical form, which restricts its suitability for datasets exhibiting high nonlinearity. Similarly, Support Vector Machine, Decision Tree, and Random Forest methods face the challenge of overfitting when applied to regression problems contrary to ANNs which have the ability to handle datasets that exhibit strong non-linear relationships between outcome variables and predictors, thereby providing more precise outcomes. Juszczyk et al. (2018) summarized advantages of using ANNs in cost estimating problems, particularly in the construction, as follows: 30 1. ANN are well-suited for regression problems where understanding the relationships between the dependent variable and numerous independent variables is challenging. 2. ANN have the capability to acquire knowledge through automated training processes, eliminating the need for extensive manual investigation. 3. ANN can build and store knowledge based on the patterns observed in real-life training examples, allowing them to learn from experience. 4. ANN exhibit the ability to generalize knowledge, enabling them to make predictions for data that were not specifically included in the training process. Neural Networks versus Conventional Methods Matel et al. (2019) stated that ANNs work more accurately than Multiple Regression Analysis (MRA) and Case-Based Reasoning system (CBR) estimating models. According to Wang et al. (2022), Artificial Neural Networks (ANNs) possess the capability to effectively handle datasets that demonstrate complex non-linear relationships between outcome variables and predictors. As a result, ANNs offer more accurate outcomes in contrast to the limitations of the regression analysis method, which relies on predetermined mathematical forms that may not be suitable for highly nonlinear datasets. Additionally, other methods such as Support Vector Machine, Decision Tree, and Random Forest encounter challenges related to overfitting when applied to regression problems. Elmousalami (2020) outlines that unlike conventional modeling methods, such as linear regression analysis, ANNs have the capability to approximate nonlinear functions with a desired level of precision. Shehatto (2013) shares differences between conventional 31 computer method and ANNs by highlighting that conventional method tackles one task at a time, with no relation or experience between each task and use cognitive approach to provide solutions. Thus, problems should be clearly identified and articulated using concise and clear instructions that leave no room for ambiguity. ANNs on the other hand do not follow a sequential or deterministic approach. They lack complex central processors and instead consist of numerous simple processors that primarily calculate the weighted sum of their inputs from other processors. In ANNs, a vast number of interconnected processing elements (neurons) operate in parallel to address a specific problem, and they have the ability to learn through examples. Unlike traditional programming, ANNs cannot be explicitly programmed to perform specific tasks. (Shehatto, 2013) Neural Network Structure According to (Najafi & Tiong, 2015), the architecture of an artificial neural network (ANN) consists of individual neurons that comprise two main components: a summing junction responsible for aggregating inputs received from neighboring neurons, and an activation function that computes the output signal, which is then transmitted to other neurons. The activation function can take various forms, including signum, linear or semi linear, hyperbolic tangent, and sigmoid functions. Neurons are organized into different layers to form a network, namely the input layer, hidden layers, and output layer (Najafi & Tiong, 2015). According to Elmousalami (2020), there are various types of activation functions that serve different purposes within artificial neural networks. These include the linear function, step function, ramp function, and tangent sigmoid function. The Selection of parameters in 34 Figure 4: The Structure of Artificial Neural Network. Extracted From (Jinisha & Jothi, 2019) Terminology Used In Artificial Neural Network The architecture and functionality of ANN are described using a particular group of terminology. For an understanding of how ANN function within themselves, it is fundamental to comprehend these terminologies. Neurons. Also referred to as nodes, are fundamental processing units. These terms are often used interchangeably. Neurons receive inputs, carry out computations, and generate an output. Weight. In the context of Artificial Neural Networks (ANNs), the term "weight" pertains to the parameters that gauge the strength of the connections linking neurons. These weights are responsible for determining how much influence each input has on the output produced by a neuron. (Chakraverty et al., 2019; Shehatto, 2013) Hidden Layer. also referred to as “Intermediary layer”, is an essential part of the structure of any network. Günaydin & Doǧan (2004) describes the hidden layer's role is to identify and retain relevant features and sub-features from the input patterns, enabling the network to make predictions about the output layer's values. 35 Learning Algorithm. learning can be defined as the method of assigning appropriate values to weights. It is generally associated with the training process; training refers to the process of attaining the desired output by adjusting the weights in the connections between layers of a network through multiple iterations. This allows the network to learn and improve its performance to accurately process information and produce the required results (Chakraverty et al., 2019). Haykin (2009) describes learning algorithms as a school process with two main categories of learning: supervised learning (learning with a teacher), and self-learning, divided into reinforced learning and unsupervised learning. Supervised Learning Method. Najafi & Tiong (2015) defined the network is provided with input and output training pairs. The sample pairs presented to the network teach it to make predictions or produce desired outputs based on the given inputs. Haykin (2009) defines the teacher, equipped with knowledge of the environment, guides the neural network using input-output examples. The network adjusts its parameters based on the training data and error signal, aiming to emulate the teacher's knowledge stored in fixed synaptic weights as long-term memory. According to Shehatto (2013), supervised learning encompasses several learning algorithms, including the Back-propagation Learning rule, Gradient Descent Learning, and Delta Rule. Among these, the Back-propagation Learning rule is the most widely used method and offers different algorithms such as Levenberg- Marquardt and Momentum for implementing the backpropagation algorithm. 36 Figure 5: Diagram of Learning With a Teacher. Extracted From (Haykin, 2009) Unsupervised Learning Method. It is where the network does not have a predefined set of categories to learn from. Instead, it independently learns and develops its own representation of the input stimuli (Haykin, 2009). Chakraverty et al. (2019) describes unsupervised training in a neural network is performed without a teacher where the target output is unknown for training the input vectors. The network adjusts weights to group similar input vectors together. Implementing unsupervised training is complex and challenging. Reinforcement Learning. It is a machine learning approach where a network learns to make decisions by interacting with an environment and receiving feedback in the form of rewards or punishments. Through a process of trial and error, the network aims to discover the optimal actions to maximize its cumulative reward (Sutton & Barto, 2018). Researchers like Mnih et al (2015) suggest reinforcement learning is often implemented using algorithms such as Q-learning, SARSA, or Deep Q- Networks (DQNs) (Mnih et al., 2015). 39 Figure 8: The Diagram of Artificial Neural Network Function: Type Sigmoid. Extracted From (Artificial Neural Network, n.d.) Piecewise Linear Functions. Bishop (2011) defines it as functions composed of multiple linear segments that are defined over distinct intervals or regions. These segments are connected at breakpoints, and the behavior of the function within each segment is determined by a linear equation. (Bishop, 2011) Figure 9: The Diagram of Artificial Neural Network Function: Piecewise Linear. Extracted From (Artificial Neural Network, n.d.) 40 Gaussian Functions. As defined by Bishop (2011), are mathematical functions that follow a bell-shaped curve with a symmetric distribution around the mean. The shape of the curve is determined by the standard deviation. These functions are commonly used in various fields, such as neural networks, for modeling continuous and smooth distributions of data. The mathematical formula for a Gaussian function is f(x) = exp(-(x - μ)² / (2σ²)), where exp denotes the exponential function, x represents the input value, μ is the mean, and σ is the standard deviation (Bishop, 2011). Figure 10: The Diagram of Artificial Neural Network Function: Gaussian. Extracted From (Artificial Neural Network, n.d.) Overfitting. Occurs when a neural network achieves high performance on the training dataset but performs poorly on new, unseen data (Bishop, 2011). Generalization. Refers to the network's ability to perform well on new, unseen data beyond the training dataset (Goodfellow et al., 2016). 41 Regularization. Encompasses methods used to prevent overfitting by imposing constraints on the weights or modifying the loss function (Chandanshive & Kambekar, 2019). Dropout. It is a specific regularization technique that randomly deactivates some neurons during training to mitigate overfitting. Normalization. It is the procedure of adjusting data to a standardized range in order to enhance the performance and convergence of an ANN. This entails rescaling the data so that it lies within a specific range, commonly 0 to 1 or -1 to 1. By normalizing the data, the dominance of features with larger scales is mitigated, promoting more balanced learning in the network. This ensures that the data is consistently formatted, enabling efficient training of the ANN (Geron, 2019). Denormalization. normalization and denormalization are data preprocessing and postprocessing techniques in ANN that aim to improve the accuracy and interpretability of the network's outputs. Denormalization is the opposite of normalization, where the normalized data is transformed back to its original scale or range. It is commonly used on the output of an ANN to obtain meaningful predictions or results in the original data scale (Rumelhart et al., 1986). Types of Artificial Neural Networks Haykin (2009) classified architecture of ANN into three distinct categories: Single Layer Feedforward Networks, Multilayer Feedforward Networks and Recurrent Networks. Elmousalami (2020) notes that the selection of the network architecture depends on the nature and requirements of the problem concerned. 44 Figure 13: Multilayer Perceptron (MLP) Showing Backpropagation Algorithm. Extracted From (Alcineide et al., 2021) Recurrent Networks. A key distinction between a recurrent neural network and a feedforward neural network is the presence of at least one feedback loop in the former (Haykin, 2009). By incorporating feedback connections, the preceding layers are able to receive data flow from the following layers (Najafi & Tiong, 2015). It is a specific type of neural network that incorporates feedback connections, enabling the flow of information from preceding layers to subsequent layers. Its purpose is to effectively handle sequential and time-dependent data (Goodfellow et al., 2016). He articulates that it is used to model relationships between sequences and other sequences rather than just fixed inputs. Figure 14: Recurrent Networks Diagram. Extracted From (Najafi & Tiong, 2015) 45 Training of Neural Network The training process is a crucial and important step in the development of ANN and the accuracy of its results. It is the process of attaining the desired output by adjusting the weights in the connections between layers of a network through multiple iterations. This allows the network to learn and improve its performance to accurately process information and produce the required results (Chakraverty et al., 2019). Shehatto (2013) articulates that the training process acquires weights meaningful information, whereas prior to training, they are initialized randomly and do not possess any specific meaning. During the training process, the weights in the neural network are initialized with small random values. For each input, the network produces an output, which initially is random. The difference between this output and the desired output, known as the correct value, is computed using one of the performance measures. The training is repeated multiple iterations setting different weight configurations. The total error of the network is obtained by summing up the squared differences over all training examples. A perfect network would have an error of zero, and a smaller error indicates a better-performing network (Krogh, 2008). Training process can be described as the following steps (Jinisha & Jothi, 2019). 1. Initialize the network: Set up the network architecture, including the number of layers, neurons, and activation functions. 2. Define the loss function: Choose an appropriate loss function that quantifies the difference between the predicted outputs and the true outputs. 3. Initialize the weights: Assign initial random values to the weights and biases of the network. 46 4. Forward propagation: Pass the input data through the network to compute the predicted outputs. 5. Compute the loss: Compare the predicted outputs with the true outputs using the defined loss function to calculate the error. 6. Backpropagation: Propagate the error backward through the network, adjusting the weights and biases using optimization algorithms like gradient descent. 7. Update the weights: Use the computed gradients to update the weights and biases, aiming to minimize the loss. 8. Iterate the forward propagation, loss calculation, backpropagation, and weight updates for a certain number of epochs or until convergence. 9. Evaluate the model: Use a separate validation or test dataset to assess the performance of the trained model. 10. (Optionally) Fine-tuning and optimization: further refine the model by adjusting hyperparameters, exploring different architectures, or employing regularization techniques to improve performance and prevent overfitting. There are various techniques used to train ANNs such as Backpropagation, Stochastic Gradient Descent (SGD), Adam Optimizer and Convolutional Neural Networks (CNN). The thesis will elucidate the first two techniques. Shehatto (2013) describes backpropagation algorithm as one of the most powerful and commonly used algorithms for training. According to (Jinisha & Jothi, 2019), Back-propagation learning algorithm was invented in 1969 for learning in multilayer network. Tijanić et al. (2020) outlines that backpropagation, the squared error between the output values and the desired values is minimized using gradient descent. The error signals 49 However, there is a risk of overfitting the validation subset with the chosen model's parameter values. To mitigate this risk, the selected model's generalization performance is measured on a distinct test set that is separate from the validation subset (Haykin, 2009). Being mindful of potential biases and pitfalls in cross-validation is crucial. Varma & Simon (2006) highlight the biases that can occur when cross-validation is used inappropriately, such as data leakage and overfitting. To mitigate these biases, they suggest important precautions, including proper randomization of data prior to applying cross- validation and exercising caution when comparing models with differing complexities (Varma & Simon, 2006). Najafi & Tiong (2015) outline that the utilization of cross- validation ensures that the neural network is not overtrained, meaning it performs well on the training set but poorly on test data. By employing this technique, a stop criterion is selected to halt training as soon as the error on the cross-validation set begins to rise. Since there are no set rules for configuring neural networks, the practitioner should construct multiple networks and select the one with the lowest error (Najafi & Tiong, 2015). Testing of Neural Network During the testing phase, the accuracy of the model's predictions is assessed. The predicted results are compared to the actual results, and the percentage error in cost estimation is computed (Günaydin & Doǧan, 2004). The author continues to describe that the impact of each network input on the network output can be examined through analysis, providing feedback on the most significant input parameters. This can be achieved through sensitivity analysis, a method that reveals the cause-and-effect relationship between inputs and outputs. By conducting sensitivity analysis, the network's size can be reduced, resulting 50 in a simpler model and shorter training times. During sensitivity analysis, network learning is disabled, ensuring that the network weights remain unaffected. The approach involves modifying the inputs to the network and observing the resulting percentage change in the output (Günaydin & Doǧan, 2004). Evaluation metrics such as accuracy, precision, recall, F1 score, or MSE are commonly used to quantify the performance of the network (Alpaydin & Bach, 2014). The allocation of data for training and testing is dependent on data availability. The percentage of cases used for testing compared to training differs between the literature. Alcimede et al. (2021) state the percentage of training to testing can range from 70/30 (70% for training and 30% for testing), and up to 90/10 (90% for training and 10% for testing). Shiha et al. (2020) defined 80/20 (80% for training and 20% for testing). According to Shehatto (2013), a smaller subset of one-tenth of the training data can be allocated for cross-validation purposes. Jinisha & Jothi (2019) outlined in their study that they used a ratio of 70% for training, 15% for cross-validation and 15% for testing. Performance Measures of ANN Model There are several performance measures used to evaluate the effectiveness of an Artificial Neural Network (ANN) model. Some commonly used performance measures in the context of ANN models include: 1. Mean Squared Error (MSE): MSE calculates the average of the squared differences between the predicted and actual values. The MSE is calculated by taking the average of the squared differences between the predicted and actual values. It gives 51 more weight to larger errors due to the squaring operation. The MSE can be expressed as: MSE = [(1/n) * Σ(y - ŷ)^2] (Chollet, 2018). where n is the number of data points, y is the actual value, and ŷ is the predicted value. It provides a measure of the overall accuracy of the model's predictions. 2. Mean Absolute Error (MAE): it is calculated by taking the average of the absolute differences between the predicted and actual values. It treats all errors equally without giving more weight to larger errors. It can be expressed as: MAE = [(1/n) * Σ|y - ŷ|] (Chollet, 2018). It provides a measure of the average magnitude of errors in the model's predictions. 3. Mean Absolute Percentage Error (MAPE): MAPE is calculated by taking the absolute difference between the actual and predicted values, dividing it by the actual value, and then multiplying by 100 to express it as a percentage. The absolute differences are then averaged across all data points to obtain the mean absolute percentage error (Hyndman & Koehler, 2006; Matel et al., 2019). Tijanić et al. (2020) defined MAPE by the formula below. 4. Root Mean Squared Error (RMSE): it is a metric used to quantify the average magnitude of prediction errors in the same units as the target variable. It is obtained 54 Sensitivity analysis involves temporarily disabling the network learning process, ensuring that the network weights remain unchanged. The fundamental concept behind sensitivity analysis is to systematically modify the input values of the network and observe the resulting changes in the output. These changes are typically quantified as percentages, providing insights into the sensitivity of the network's output to variations in the input values (Günaydin & Doǧan, 2004). According to Hagan et al. (2014), once the training is completed for a multilayer network, it is valuable to evaluate the significance of each element within the input vector. Identifying low-impact (unimportant) elements allows for their elimination, simplifying the network, reducing computational requirements, and mitigating the risk of overfitting. Conducting a sensitivity analysis can be beneficial when there is no definitive method for determining the importance of each input. Artificial Neural Network Properties Heravi & Eslamdoost (2015) describes ANN in construction network through immense literature background. They outlined insights into the potential applications of neural network architectures in construction engineering and management, highlighting their effectiveness as a management tool for automation in the construction industry. They also highlighted utilization and applicability of artificial neural networks (ANNs) in civil engineering where they stated ANNs function and demonstrated their effectiveness in solving various civil engineering problems. One of the studies in his report developed a 55 neural network model specifically for estimating productivity in excavation-hauling operations. Another study focused on estimating construction productivity for concrete formwork elements, such as slabs, walls, and columns. They explored different neural network structures and found that a three-layered network with a fuzzy output structure provided the most suitable model, considering the subjective nature of the input variables.(Heravi & Eslamdoost, 2015) Vikas et al. (2011) characterizes the learning process of a neural network by three main events. First, the network is stimulated by environmental input. Second, the network's free parameters are adjusted in response to this stimulation, allowing it to adapt to external stimuli. Finally, the network exhibits a modified response to the environment as a result of the changes that have occurred in its internal structure and function. Hashemi et al. (2020) highlighted the percentage of different methods applied to construction types. They outlined ANN being used in higher percentage in all types of construction projects compared to other methods of neural network such as fuzzy neural network, Support Vector Machine (SVM), Radial Basis Functions (RBF), and other methods. ANN varies from 35% in road and infrastructure projects, 40% in building projects and can reach up to 100% in water-related construction projects (Hashemi et al., 2020). ANN Problems and Challenges There are several problems and challenges associated with using Artificial Neural Networks (ANNs) in various applications. Here are some common ones: 56 • Overfitting: ANN can be prone to overfitting; it is a significant issue that can arise during the training of a neural network. It occurs when the network becomes too specialized in the training data and fails to generalize well to new situations, leading to reduced performance (Chandanshive & Kambekar, 2019). • Training time and computational complexity: Training ANNs can be time- consuming and computationally intensive, especially for large and complex networks with a large number of parameters. This can lead to longer training times and higher resource requirements (Chen & Lin, 2014). • Selection of Appropriate Architecture: Designing an optimal architecture for an artificial neural network (ANN) involves various challenges, such as determining the appropriate number of layers, neurons, and connectivity. The optimal architecture may vary depending on the specific problem and dataset (Haykin, 2009). • Data Limitations: The performance of a model is primarily determined by the quality and quantity of training data. The amount of data required for a machine- learning learning algorithm depends on the complexity of the problem and the chosen algorithm. Limited or biased data can lead to poor performance and biased predictions (Matel et al., 2019). • Interpretability: ANNs are commonly regarded as black box models, which implies that comprehending and interpreting the rationale behind their predictions can be difficult. This lack of interpretability may impose constraints on their usability in certain fields (Guidotti et al., 2018; Hashemi et al., 2020). 59 characteristics and overlook the influence of external economic factors. The study aims to quantitatively explore the effects of economic factors on construction cost estimation by using Deep Neural Networks (DNN) as an estimator and Shapley Additive explanations (SHAP) as a model interpreter. The analysis utilized data set and included a comparison analysis with other popular machine learning models used in construction cost estimation. The results indicate that economic factors play a significant role in reducing estimation errors and may even be more influential than project characteristics. The findings have practical implications for stakeholders in the construction engineering and management field, providing insights for decision-making, and contribute to a better understanding of the impact of various influential factors on construction cost estimation. The studies reviewed, including those mentioned previously, indicate that employing artificial neural networks for early cost estimation in construction projects has significant potential. These findings emphasize the importance of conducting additional research in this field to investigate and improve the utilization of neural networks for estimating construction costs at the initial project phases. 60 CHAPTER 3 Research Methodology Chapter Introduction This chapter outlines the research methodology utilized in this thesis. It begins by discussing the research strategy and provides a visual representation of the research design through a flow chart. The research design involves an in-depth analysis of recent literature to identify key factors influencing cost estimation in building projects. Subsequently, data collection is conducted through surveys to establish a correlation between these factors and the cost of projects. The collected data is then analyzed using methods presented and explained in this chapter. The end result can be used to enhance the input parameters for the ANN model that will be constructed in the next phase of the study. The model will be employed at the new projects’ early stage to produce accurate estimates of costs. Research Strategy The research strategy in this study consists of the following points: 1. Research Approach: The research approach is generally either quantitative, qualitative, or mixed methods. This research is based on mixed methods which allows for triangulation and deeper insights into research questions. Close-ended survey questions with predetermined choices are considered quantitative data that can be evaluated statistically. Afterwards, tools for development of Data collection are used such as interviews with expert reviews. They are considered a qualitative 61 research method as they entail having candid, in-depth discussions with individuals/organizations in order to assemble comprehensive, specific, and individualized data. In-depth exploration of organizational experiences, perceptions, and insights is made possible by the probing and follow-up questions that can be asked during interviews. 2. Research Design: The research design is well explained in the next paragraph where figure 15 depicts the design of every step of the research. It is based on exploratory and correlational data, where new factors are being determined while the previous literatures are used to correlate the data to close the knowledge gap in order to provide a better understanding of inputs for building exact cost estimation model. 3. Data Collection Methods: The first data collection was based on the analysis of literature reviews which were focused on least investigated factors in building ANN models for cost estimation. Afterwards, a quantitative survey is conducted with a sample of almost ninety organizations. Subsequently, tools for development of Data collection are used to improve survey results as described in the next point. 4. Sampling Strategy: The targeted population for the surveys were chosen based on their profiles and not distributed randomly to reduce misleading results. The population consisted of almost ninety organizations ranging from owners, consultants, and contractors in order to view results from all sides. The population itself was from various job positions but focusing primarily on managerial positions with a high experience to optimum results. 5. Data Analysis Techniques: Data analysis for quantitative surveys generates data that can be analyzed using statistical analysis, enabling numerical examination of 64 3. Data Collection and Development of Data Collection Tools: A structured survey was distributed to the initial group of participants, while closely noting their comments on the survey. The feedback was integrated into the survey before being distributed to the general population. After conducting the survey, some participants were chosen to answer the survey again and their answers were noted down to check reliability of answers (Test-Retest Reliability). 4. Afterwards, the next step in data collection was to conduct expert review method (interviews), to gain closer view on the main parameters affecting cost estimation procedure of building projects in UAE. The participants chosen to perform interviews were senior participants in specialized fields such as: technical manager, estimation, contracts management, planning and general manager. 5. Data Analysis: After analyzing the data collected from surveys and interviews, the data was analyzed statistically at first and interview results were used to confirm the data and draw recommendations and conclusion from the study. 6. Conclusion and recommendation phase: In this stage, the content of the thesis was written, and the research chapters were covered. Moreover, the research was summarized in the conclusion section with many recommendations. Literature Studies Numerous researchers have examined different parameters and developed their models based on diverse factors. Most of the factors can be classified into three categories: Structural design factors, finishing factors and special circumstances factors such as market index, type of client, type of contract, project delivery method, etc.... 65 The presented study focused on structural design factors and special factors as described in Table 1 below. The table provides a description of each factor along with the corresponding reference from which it was extracted. Table 1: Key Influential Factors Examined in Previous Research Number Factor Highlighted Previous Literature Reference 1 Project Size (Elfaki et al., 2014) 2 Type of Project (Elfaki et al., 2014) 3 Soil type/ Conditions • (Elfaki et al., 2014) • (Chandanshive & Kambekar, 2019) 4 Type of Client (Elfaki et al., 2014) 5 Type of Contract (Elfaki et al., 2014) 6 Building Total Area • (Günaydin & Doǧan, 2004) • (Chandanshive & Kambekar, 2019) • (Juszczyk et al., 2018) 7 Number of Floors • (Günaydin & Doǧan, 2004) • (Chandanshive & Kambekar, 2019) 8 Foundation Type • (Günaydin & Doǧan, 2004) • (Chandanshive & Kambekar, 2019) 9 Structure/ Slab Type • (Günaydin & Doǧan, 2004) • (Chandanshive & Kambekar, 2019) 10 Earthquake zone (Chandanshive & Kambekar, 2019) 66 11 Weather Conditions (Najafi & Tiong, 2015) 12 Management Conditions (Najafi & Tiong, 2015) 13 Skilled Labor Availability (Najafi & Tiong, 2015) 14 Project Location (Juszczyk et al., 2018) 15 Construction Trend value such as Fluctuation (Wang et al., 2022) Tools for Development of Data Collection Tools and methods used to develop surveys results typically include: 1. Pilot Testing: Administering the survey to a small sample group to identify any issues with the survey design, clarity of questions, and response options. 2. Test-Retest Reliability: Administering the survey to a group of participants at two different time points to assess the consistency of responses. 3. Expert Review (interviews): Seeking feedback from subject matter experts or experienced researchers to evaluate the survey's content validity, clarity, and appropriateness. Conducting interviews can be considered a data analysis method. Interviews allow for in-depth exploration of the participants' perspectives and can provide valuable insights and clarification regarding their survey responses. Interviews can help test survey findings by verifying and complementing the quantitative data collected through surveys. The information obtained from interviews can also contribute to a richer understanding of the research topic and enhance the validity of the study's findings. 69 Figure 17: Population Positions in the Construction Field. (Total Sample of 87) Participants Years of Experience. The population years of experience varies from 2 years to over 20 years in the construction field. The study also provides a view on the years of experience by market. As shown in figure 18, most of feedback are from middle east market and this is because the study is focused on UAE market. The American construction market, European market, and Asian market have minor presence in responders experience with total absence of Australian market. The highest percentage of feedback was ranging from 11-20 years with 39% of population, followed by range 6-10 years (25%) and then range of over 20 years (23%). 70 Figure 18: Population Years of Experience in the Construction Field by Market. (Total Sample of 87) Factors Affecting Cost of the Building Projects Theoretical key factors have been determined from past literatures, other factors that have been identified through expert recommendations from past experiences and both were used to build the survey. Factors of cost estimation are the main component in an ANN model where they are the input layer of the model. Inaccurate estimation can be noted as one of reasons for project cost overruns. Figure 19 shows the percentage of survey replies rating if inaccurate estimation is a reason for cost overruns in projects, where 87% of replies confirmed that inaccurate estimation is a main reason for project cost overrun. 71 Figure 19: Percentage Rating If Inaccurate Estimation Can Be Identified As Reason for Cost Overruns in Construction Projects (Total Sample of 87). Figure 20 illustrates total replies stating percentage of projects in the company with cost overrun due to inaccurate estimation. It shows that 21% of replies state that only 1/10 of projects face overruns, while 32% of the replies state that a quarter of projects face overruns. 34% of the population replied that half or more than half of projects face overruns due to the same issue. Figure 20: Population Rating Percentage of Projects With Cost Overruns Issues Due to Inaccurate Estimation (Total Sample of 87). ercentage of Replies Rating If Inaccurate Estimation Can Be a Reason for Cost Overruns in Construction rojects No es Not Sure 21 2 1 1 1 0 10 1 20 2 0 10 of projects 2 of projects 0 of projects More than 0 of projects Not Sure ercentage of Organi ation rojects with cost overrun 74 Ty S M (S , C , , …). The type of structure material for the construction of the project has a major impact on project planning, type of equipment, skilled team and labors and risk. For example, concrete reinforced projects will have longer project span, but it is a proven science with low risks and low requirement for skilled professionals and special equipment. Steel structure have relatively shorter time frame but require skilled labor which will change the cost estimations. Precast projects are fast executing and have low risk and overruns because of their factory environment, but they have a higher cost and require higher logistics and special equipment. 70% of the replies rated type of structure as moderate to high, with 42% showing high impact as depicted in Pie chart in figure 22. Figure 22: Population Rating Impact of Type of Building Structure Material Factor on Cost Estimation (Total Sample of 87). Area of Typical Floor. The area of the projects’ ground impact on cost estimation is rated almost equally from low to high as it depends on the company scale. Small companies will consider its impact higher as they will require to procure material and equipment, 10 28 42 14 Impact of Type of Structure Material (Steel, Concrete, restressed concrete, recast) on Building Cost Estimation ery ow ow Medium High ery High 75 recruit new people, and engage sometimes in jobs they have never executed before. Bigger companies have different type of consideration as bigger scale project decrease the overhead and indirect cost per square meter (better ratio of people to area of project leads to less cost of staff, as for example, a 1000-metersquare will require one engineer while a 100-meter square will also still require an engineer, but overhead will increase in the latter). As shown in figure 23, 30% of replies were high, followed by 27% as moderate and 25% as low impact; showing that its impact varies by other related conditions. Figure 23: Population Rating Impact of Floor Area Factor on Cost Estimation (Total Sample of 87). Number of Floors. The number of floors has a different perspective if it can be considered a major factor or not. If the number of floors is small, then, the impact difference between Ground and five levels (G+5) or a G+10 is low, other than the quantities that are considered. When the project becomes a high-rise, the overhead and indirect cost will rise, and the factor becomes high rated because manlifts will be added, customized special cranes and concrete might be considered, plumbing and 2 2 0 1 Impact of Area of Typical Floor on Building Cost Estimation ery ow ow Medium High ery High 76 sewer systems needs to be created at certain levels for crew needs, risks are higher and will require more expensive insurances, higher safety requirements, crew training, etc.… Replies were distributed mainly from moderate to very high with highest rating was “High” with 1 of replies followed by “ ery high” and “Moderate” at 2 as shown in figure 24. Figure 24: Population Rating Impact of Number of Floors Factor on Cost Estimation (Total Sample of 87). Ty S b (S , R bb …). The type of slab is generally related to the factor stated previously “Type of Structure Material”. Its impact will not be important comparing main structure material, hence, the replies rating the factor were distributed from low to moderate to high with 17%, 32% and 29% respectively as illustrated in figure 25. As stated previously, the impact will vary on size of organization and availability of material and equipment, usage of subcontractor and many other factors. 1 2 1 2 Impact of Number of Floors on Building Cost Estimation ery ow ow Medium High ery High 79 shown in figure 28 below. Its effect from expert review perspective is minor and can be defined as “ ery low”. Third of the replies was “Moderate” with quarter of replies rating “ ow” and another quarter rating “High”. Figure 28: Population Rating Impact of Type of Contract Factor on Cost Estimation (Total Sample of 87). Area of Shear Walls in Project. Area of shear wall depend on three factors mainly: availability of formwork and labor, risk of formwork opening, and time frame for execution. Third of the survey (30%) population rated it moderate which gives it accurate description, and half of replies was divided between low and high as shown in figure 29. Experienced people classified as low, operation people and unexperienced opted for higher classification. 11 2 2 Impact of ength of Columns Spans on Building Cost Estimation ery ow ow Medium High ery High 80 Figure 29: Population Rating Impact of Area of Shear Walls Factor on Cost Estimation (Total Sample of 87). Location of Project (Country, City). The location of the project can be classified in the top five in terms of impact on the cost estimation. It is linked to most of the other important factors and its impact will consist of the below points to be able to provide a decent cost estimate. • Is the project location in an urban, suburb or in the city? • Is it located in an isolated area or near material sources? • Is the location previously recognized or new to the company? • How’s the political situation and customs and regulations if foreign country? All the factors above will contribute to a noticeable rise in cost because of unknown risk and in procurement of insurances to mitigate risks. It can include overhead and indirect costs such as staff housing if isolated area, challenges in transportation, might require building of temporary roads and service networks, raw materials (gravel, cement, sand, and others) might need higher time to deliver or need to look other suppliers with higher prices, area labor might be scarcer, etc. Figure 30 shows 1 2 0 2 Impact of Area of Shear alls on Building Cost Estimation ery ow ow Medium High ery High 81 34% of replies were “Moderate” followed by 0 as “High” and another 0 divided between “ ery high” and “ ow”. Figure 30: Population Rating Impact of Location of Project Factor on Cost Estimation (Total Sample of 87). Usage of Building. The usage of building factor can be classified into two main categories: governmental (public) or private project. Governmental projects will have a higher standard, regulations, and Leed requirements, opposed to private where the budget will determine the rest. Subcategories such as commercial, schools, residential or facility can further impact the estimation, but the main impact will always whether governmental or private. Figure 31 below shows that 32% of replies were “High”, followed by 29 “Moderate”, and 1 as “ ow”. 1 4 0 1 Impact of roject ocation on Building Cost Estimation ery ow ow Medium High ery High 84 B), if the contract is lumpsum, if the contract conditions are imbalanced. One example is a construction in Abu Dhabi, UAE where the project where D-B and the soil report was not ready at the time of estimation; the project land was full of cavities and concrete volume for piles were underestimated. The contract was a fixed lump sum causing a loss of profit for the contractor. Figure 34 states that 32% of replies were “high”, followed by 2 “Moderate” and 22 as “High”. Figure 34: Population Rating Impact of Type of Soil Factor on Cost Estimation (Total Sample of 87). Complexity of the Project. Complexity of project was rated highest impact on cost estimation. As can be noticed in Figure 35, 40 rated as “ ery high”, followed by 1 as “High”, then 20 as “Moderate”, building total percentage to 90 as moderate or higher. The main reason for this high percentage of inaccurate estimation analyzed from the study led from complexity of drawings and because of the low experience of some of the estimation team, quantities may be estimated wrong or even drawings may not be understood. Complex projects may also require new execution methods 4 10 2 2 22 Impact of Soil Type on Building Cost Estimation ery ow ow Medium High ery High 85 that are hard to predict cost, higher non-mitigable risks, etc. This is a conclusion of the study, but complexity as a factor is a wide subject where the factor can be better defined, and more conclusions can be withdrawn from deeper studies on the factor impact on the cost estimate. Figure 35: Population Rating Impact of Project Complexity Factor on Cost Estimation (Total Sample of 87). Usage of Building Information Modelling (BIM). The usage of advanced techniques such as Bim has improved the quality of cost estimation process by providing accurate automatic quantity survey for different material and enhance the understanding of complex projects. It also improved the coordination between different trades, enabling to detect clashes before happening which decreased abortive work. Figure 36 shows that more than 76% of replies stated that they use BIM in half or more in their project. this proves that it is widely used. Dubai projects, in the UAE, are now obliged to use BIM for the handover as a requirement from the client. 2 20 1 40 Impact of roject Complexity on Building Cost Estimation ery ow ow Medium High ery High 86 Figure 36: Population Rating Their Organization Usage of Advances Techniques (I.E., BIM) (Total Sample of 87). Figure 37 highlights the importance of usage of BIM as 34% where replies gave a moderate rating on impact with 25% rating impact high and 20% rating very high. This means that 80% of replies agreed impact vary from moderate to high. Figure 37: Population Rating Usage of BIM Impact on Cost Estimation (Total Sample of 87). Weather Conditions (Severe Fog, Sandstorm, Rainfall). The impact of weather conditions on cost estimation varies on the case presented. Some cases can be extreme and unknown, and others can be mitigated and known no matter how extreme the case is. For example, marine projects are 8 1 2 0 10 20 0 40 0 0 10 of projects 2 of projects 0 of projects More than 0 of projects ( M, ) Companies usage percentage of advances techniques (i.e. BIM, Advanced estimation softwares) 1 4 2 20 Impact of sage of Building Information Modelling (BIM) on Building Cost Estimation ery ow ow Medium High ery High