“Integrated Approach Of Cost Estimation For Road Projects By Using Artificial Neural Network (Ann) And Analytic Hierarchy Process (Ahp)

Abstract

Initial decisions on a construction project have a significant influence on upcoming project performance. The process of developing comprehensive project cost estimation is serious for a project to be effective on completion. Cost estimation is being mentioned as the procedure of examining a specific scope of work and estimating the cost of completing the work. The main objective of this research is to develop AHP-ANN model to estimate the cost at early stage of road projects in Ethiopia using parametric techniques & minimize the percentage error of estimation. Torealize this there is a requirement to identify the factors that affect the cost of road projects that can be available at early stage. There are so many factors that affect project costs in the early stages of project construction are still unspecified well. Therefore, the issues of this study to identify and evaluate factors affecting the accuracy of cost estimates and provided best models to arrive at a better and reliable cost estimation and to support decision-makers in predicting cost of road projects during construction period to control lack of budget in Ethiopia. Initially, the study starts from AHP defining the alternatives that needs to be evaluated in this case 47 different types of road projects historical data obtained from 2012- 2019 executed projects all over Ethiopia except the capital city Addis Ababa. Then identified the Problem and selected criteria then established priority between criteria using pairwise comparison and check consistency for the set of judgments then after determined the relative weights using mathematical calculation based on the data and assigned relative weights to the criteria. It represented an accurate approach to quantifying the weights of magnitudes of factors through pair wise comparisons. Each of the respondents compared the relative importance each pair of items using a specially designed questionnaire or interview. The next plan of the study was to develop ANNs model. The model was a code-based that generated the ANN using MATLAB® and EXCEL for the calculation to simplify its use. Configured the network’s inputs and outputs in this case seven input parameters considered to train the neural network including each criterion combined weight plus weighted ranking based on calculated weight. target/output was total actual cost of the sampled road projects then adjusted the network parameters (the weights and biases) to optimize performance separated randomly the total data set in to three for training, validation and test set. The study provided three Models by varying the portion of the parameters from set data. On the way to trained the network and validated the network’s results the study used FF or MLPs of networks with BP method applied to resolve the problem much more powerfully and to approximate some function, also exhausting LM Algorithm to minimizing the MSE of a NN and most common predictive modelling techniques LR. The architectures of ANN were adopted after several trials and error. The best total accuracy performances of Model I, 99.89%, 99.83% and 99.42%, Model II, 99.99%, 99.86% and 99.76%, Model III, 99.75%, 99.80% and 99.6% for training, validation and test set respectively. This research was accomplished the capability to estimate the cost of road projects at early-stage developed certified models and determined the factors that highly affected the cost estimation of road projects during feasibility and planning stage and reduce the percentage error of cost estimation in Ethiopia. Therefore, the results are expected to provide better project cost prediction system with lowest error that can be used by society for planning and application of road construction projects.