The past literature review shows that the ANN is quite efficient in prediction of tender price and prediction of tender cost [18, 37, 60, 64, 87, 95, 96, 98], cash expenses in a project [25, 75, 99], labor costs [76, 82], prequalification of contractors [58], performance of contract [96], and quantification of risk [64].


The cost estimation is required for proficient functioning of the construction industry. The past survey of the literature indicates a comprehensive usage of ANN in determination of construction cost [16,18,23,29, 56, 74, 88, 87, 94].

Williams [98] was one of the first researchers who forecasted variation in the construction cost for six months and confirmed the efficiency of the BP algorithm. Hegazy et al. [51] observed that the ANN performs efficiently in different fields of construction management. Hegazy and Ayed [50] adopted the genetic algorithm and optimization techniques for cost optimization of highway projects. Adeli [2] developed an ANN model for optimizing the cost of reinforced concrete (RC) pavements. The analysis results showed that parameters such as atmospheric conditions and human judgment considerably influence the cost forecasting. The authors further observed overfitting to be one of the main problems in the ANN models. Geiger [41] used the ANN network for cost estimation of sheet metals. The authors developed separate models for cost estimation of material, power endurance, and accessories. The results showed that ANN models observed accuracy ranging from 85% to 95%. Elhag and Bossabine [34] proposed the ANN model for cost estimation of buildings.

The results showed an efficient performance of ANN models with high accuracy. Al-Tabtabai [13] determined an incremental increase in the cost of the construction project using ANN and confirmed considerable accuracy of ANN models. Bhoklia and Ogunlana [23] developed an ANN model forecasting the building cost by the inclusion of variables such as finishing, structural system, the height of the building, type of decoration, accessibility to the site, etc. The analysis results showed that the proposed model underestimated the values for 42.7% samples and overestimated the results for 57.3% samples. Fang and Froese [38] developed an ANN model and predicted the relationship between parameters such as concrete cost, type of formwork, and quality of concrete used in high-performance concrete. Some researchers advocated the use of hybrid networks in predicting the cost estimate for the buildings. However, the comparison showed the poor performance of the hybrid models as compared to the ANN model. Shtub and Versano [86] used both regression and neural network techniques in determination of steel pipe cost. The authors observed the better performance of the linear regression model as compared to ANN, which was in contradiction to the results of the previous literature study. Assaf [17] observed that ANNs predict the overall effective cost of construction projects in comparison to traditional models. Emsley et al. [37] used a dataset of 288 properties to determine the efficiency of both regression and ANN models. In their research study, the authors used a large number of independent variables, including design and site-related variables. The maximum mean absolute percentage error was observed as 17%, which was veiy large and cannot be prescribed in practical applications.

Setyawati et al. [83] proposed an ANN model for cost estimation of institutional buildings. The results of the analytical study confirmed a high prediction accuracy of the model. Pathak and Agaiwal [73] observed ANN to accurately predict the construction cost of water tanks made of reinforced cement concrete. The authors adopted input parameters like column number, height to diameter ratio, conical wall angle, etc., for modeling ANN. The analysis result showed that the proposed network predicted the cost of the water tank with high accuracy. Giinaydm and Dogan [48] advocated the utilization of the ANN model for prediction of the construction cost of buildings. In the proposed ANN model, building height, width, and length were used as the input parameters. The proposed ANN model predicted the results with high accuracy and was easily applicable in the initial project phase. Sonmez [88] studied the influence of parameters such as location, car parking area, and common area as independent variables. The authors compared the linear regression model with a neural network and observed neural network to exhibit a better accuracy of 88%. Gtinaydm and Dogan [48] developed an ANN model and predicted the cost of building projects with an accuracy of 93%. The authors confirmed that adoption of neural networks reduced the uncertainty in the estimation of building cost. Kim [56] determined the costs of buildings located in Korea and observed the superior performance of the ANN model. Wilmot and Mei [99] reported observed high efficiency of ANN models in separately predicting the labor and construction cost. Sodikov [87] used backpropagation ANN for cost estimation of highway projects and found its accurate prediction. Sayed and Iranmanesh (2008) developed an ANN model to reduce the risk of an increase in project cost and observed its satisfactory performance in comparison to traditional methods. Bouabaz and Hamami [24] observed that the ANN model predicted the maintenance and repair cost of the bridges accurately. Sonmez and Ontepeli [89] observed the superior performance of ANN in comparison to the regression model in the cost estimation of an urban railway system. Wang et al. [94] utilized the BP algorithm for cost estimation of highway projects. The authors trained the model based on a large number of datasets. The authors observed that BP algorithm predicted the cost with considerable accuracy. Arafa and Alqedra [ 16] developed an ANN model, which was trained based on the experimental dataset of buildings. The analysis results showed that area of the ground floor, type of foundation, and exterior surface area to be the parameters had a strong influence on the preliminary building cost. Elsawy and Higgins [36] developed an ANN model trained based on the experimental dataset of 52 real-life projects constructed from 2002 to 2009 in Egypt. The authors observed that the ANN model yielded an accuracy of 80% in the assessment of the construction cost.

Ahiaga-Dagbui and Smith [9] developed a neural network model using the experimental dataset of building projects in Scotland. The authors developed different models for normalizing the target cost, weights, and cost variable transformations to predict the acmal cost. The analysis results showed that the ANN model to be efficient in the prediction of the effective construction cost. Ebrahinmejad et al. [32] proposed a model based on the concept of support vector machine (SVM) for the estimation of construction cost during the preliminary stage. The proposed model exhibited a close correlation in comparison with nonlinear regression and BP algorithm. Alqahtani and Whyte [11] developed an ANN model based on a dataset of 20 building models. The model computed the cost of significant and insignificant items and observed high accuracy. Lyne and Maximino [62] developed an ANN model and used an experimental dataset of a large number of buildings to train the model for predicting the total cost of structural members. The authors observed that results yielded by the proposed model were in close agreement with the experimental results. El-Sawah and Moselhi [35] showed that the mean error of a neural network to range from 17% to 20% for the ANN model used in the proposed study. In addition, the authors observed a linear regression model to be primarily influenced by the training data in comparison to ANN. Yadav et al. [102] proposed an ANN model and trained it based on the experimental dataset of past two decades to predict the cost of the building projects in the preliminary phase. The analysis results showed a high accuracy of the proposed model in comparison to the traditional techniques.

Alshamrani [12] developed a multilinear regression model to compute the initial and sustainable cost for RC and steel framed buildings. The input variables adopted in the model were building area, the height of the floor, and the number of levels. The proposed model was observed to yield accurate results in comparison to traditional techniques. Jeong et al. [53] developed a system to develop building energy efficiency. The authors developed the prediction model using data mining techniques based on a dataset of 437 building models. The authors conducted validation studies on multistory buildings. The results of the analytical research indicated high accuracy of the proposed model.

Bayram [20] conducted a research study to validate “Bromilow’s time- cost model” for the estimation of project duration. The ANN model was developed and trained based on the experimental dataset of a large number of buildings. The authors adopted the input data such as the total area of construction, height of the building, gross floor area, actual cost, and project duration. The analytical studies demonstrated the accuracy of the proposed approach. Zhang et al. [105] aimed at improving the efficiency of predictions of cost forecast through a time-series approach. The results indicate the accuracy of the proposed plan in comparison with other methods in project scheduling and cost.

Juszczyk et al. [54] proposed an ANN model for estimation of the building overhead cost. The proposed model was trained based on a dataset of 143 buildings using multilayer perception theory with varying activation functions. The analysis results showed that the proposed model exhibited satisfactory results in comparison to the previous models. Kang and Kim [55] used ANN for risk assessment studies. The proposed model was developed in two steps. First, the risk information is collected. Based on the information, a unique software program was developed. The second stage consisted of developing a case study by analyzing the risk information for two plant projects. The analysis results showed that the proposed model exhibited greater accuracy in comparison to the models adopted by the previous researchers.

Xu et al. [101] predicted the cost inclined due to damage inclined during seismic excitation. The proposed model was based on Federal Emergency Management Agency (FEMA) approach, and BIM software was used for the analytical study. In the first stage, tune history analysis was conducted, and an algorithm was proposed to identify the damage. After that, fragility curves were developed subsequently. In the second stage, the BIM model was prepared based on the building configuration. In the final step, an algorithm to visualize the component was designed. The proposed algorithm was used to derive results for a six-stoiy building in Beijing. The analysis showed a high accuracy of the proposed algorithm.


Construction engineering has integrated components of planning, designing, management, and construction of civil engineering structures such as buildings, bridges, airports, highways, railways, dams, etc. The past literature survey shows applications of ANN in decision making, risk prediction, project scheduling, optimization, resource allocation, etc. Moselhi et al. (1991) have enumerated the applications of ANN in the construction sector. Boussabaine and Kaka [25] conducted a literature survey on the application of ANN in construction management.

Adeli and Karim [5] used ANN in construction scheduling of highway projects. The ANN was employed for solving nonlinear problems of construction scheduling, which involved minimization of the project duration. The proposed modal incorporated the features of both linear scheduling and critical path method for optimization of construction cost. The proposed methodology was observed to be more efficient as compared to traditional methods. Graham and Thomas [47] applied ANN for predicting manufacture time of ready mixed concrete, which was sensitive to the construction operations. The authors adopted the input parameters such as operation time, the month of operation, the total volume of operation, volume of the truck, load arrival, average interval time, etc.

Yahia et al. [103] utilized ANN for prediction of project duration time and observed superior performance and simple application of ANN as compared to traditional methods. Petruseva et al. (2013) proposed a supervised learning algorithm termed as SYM for forecasting the duration of the construction project. The ANN was trained based on the dataset of 75 construction projects completed between 1999 and 2011. The studied projects were located in Bosnia and were obtained through field survey and analysis. The authors also used regression analysis to predict the construction duration and compared it with ANN. The analysis results showed that ANN yielded accurate results as compared to regression analysis.

Maghrebi et al. [63] utilized the ANN in predicting the manufacturing duration of ready mix concrete. The authors mainly focused on supply chain parameters concerned with ready mix concrete. The model was proposed and calibrated based on the experimental dataset of buildings located in Sydney. The comparison of ANN with traditional methods showed its superior performance.

Bhargava et al. [21 ] proposed a new model to overcome the disadvantages of previously proposed models. The authors developed a technique called a line of balance which was applicable to both linear and repetitive works. Based on the analysis work, the authors suggested that there is a need for robust theory to address the critical aspects of planning and scheduling aspects. El-Gohary et al. [33] introduced a new ANN model to predict the cost of labor in a construction project. In developing the model, a wide range of influencing parameters such as working hours and the number of labors were considered. The proposed approach was applied to model the labor productivity of foundation work. The recommended results showed adequate convergence and accuracy. Nevertheless, the authors proposed new activation functions to achieve better results. Alaloul et al. [10] used multilayered feedforward ANN in predicting the project cost and project duration. The proposed model showed accurate results in comparison with other existing methods, with an average mean square value of 0.0231. Andersen and Findsen [15] developed a pragmatic design strategy based on design and the О and M phases. The proposed approach and implemented and tested for a variety of building models to get accurate results. Ballesteros-Perez et al. [19] proposed two nonlinear theoretical models to estimate the construction cost. The main advantage of this model was that they were both discrete and continues. Moreover, it was observed that the proposed model could be applied to a variety of construction projects. The authors validated the model by testing it on several building models. In addition, the proposed models were very efficient in predicting the crashing cost of the project. Similar observations were reported by Ferreira et al. [39].


The ANN models have been widely utilized in construction management. The neural networks have been increasingly used to determine the construction cost [37, 50, 98] and to predict the tender bids [60, 64]. for monitoring the budget of construction [30]. construction demand [43], cash flow in a project [25], earthmoving operation (Shi, 1999), labor productivity (Savin and Fazio, 1998; [27, 76], effectiveness (Sinha and Mckim, 2001), new technology development (Chao and Skibniewski, 1995), and organizational behavior. Hegazy et al. [51] developed a neural network for optimization of construction activities and cost. Chua et al. [30] developed an ANN model for identification of the factors affecting the construction budget and situations associated with it. The authors obtained the data from field survey to develop a budget performance model. The proposed approach allowed the model to build the relationship between input and output parameters. The authors used eight input parameters such as project team, planning and control efforts, project manager, hierarchy level in the organization, construction duration, design parameters, and frequency of team meetings. The proposed model was capable of yielding good prediction even with unseen data. The model was observed to perform with excellence in different aspects of construction management.

Assaf [17] proposed an ANN model for determination of overhead project cost in Saudi Arabia. The analysis results showed that the material and project duration significantly influenced the completion time of the project. The authors showed that negative effects of unstable construction on overall project cost.

Emsley et al. [37] proposed an ANN model, which was framed based on the dataset of 300 building projects. The analysis was conducted using both the ANN and the linear regression model. The analysis results showed better performance of ANN in comparison to linear regression analysis. Nevertheless, the ANN was found more efficient in case of nonlinear problems as compared to linear regression analysis. Gtinaydin and Dogan [48] developed an ANN model for determining construction cost of beams and columns in an RC framed building. The authors trained the ANN building model using an experimental dataset of 30 buildings and adopted eight input parameters, which were a function of building properties. The analysis results showed that the proposed ANN model exhibited a close correlation with the experimental results.

Kim [56] advocated the use of hybrid models by combining the effects of the backpropagation algorithm and genetic algorithms in cost estimation. The authors developed an ANN model and trained it based on data of 530 buildings located in Korea during the period between 1997 and 2000. The analytical study showed better efficiency of the hybrid algorithm as compared to the ANN model. This study paved the way for the adoption of such hybrid algorithms.

Wilmot and Mei [99] developed a cost index describing the overall construction cost of highways. The proposed index included parameters such as cost of construction, cost of labor, and equipment. The analysis results showed that the proposed model yielded accurate results and was veiy good at forecasting the future prices of the highway projects. Sodikov [87] improved this research work and proposed a new ANN model by including the uncertainties in the project during the design phase. The proposed model was observed to be more accurate in comparison to the previously proposed models, which ignored the uncertainties involved in the project cost.

Golpayegani and Emamizadeh [46] used ANN for the breakdown of the construction activity into several parts and rescheduling it to avoid unnecessary delays in the construction duration. The proposed model was observed to be more accurate in comparison to the existing models in reducing the uncertainties causing project delays. Naderpour et al. [70] developed an ANN model to forecast the project cost. The developed system was based on earned value management systems based on projects selected on a random basis. The proposed model was compared with the previous approach, which showed the efficiency of the previous procedure. Wang et al. (2009) proposed two building models based on bootstrap technology for pre and post planning of buildings. The results demonstrated the efficiency of the proposed approach. Elsawy and Higgins [36] adapted an ANN approach for the development of a cost estimation model to determine overhead site cost for buildings in Egypt. The authors used 52 real cases to train the neural network model.

< Prev   CONTENTS   Source   Next >