REGRESSION AND BACK PROPAGATION NEURAL NETWORK MODELS FOR PREDICTING HIGHWAY CONSTRUCTION DURATION IN NIGERIA
Publication Date : 01/08/2017
Reliable prediction of construction duration at the planning phase is essential for project feasibility studies, budgeting decisions, project monitoring and performance evaluation. This research examined the performances of parametric models for the prediction of duration of highway construction projects. Three generalized linear regression models in the form of linear, semi log and log-log and a Back Propagation Neural Network (BP-NN) model were developed using a dataset of 57 successfully completed highway projects. 80% of the data were used for developing the models while the remaining 20% were used as test samples for validating the models. The results of the regression analysis revealed that the linear, semi-log and log-log models have R2 values of 0.546, 0.631 and 0.940 respectively indicating a good fit to the data in all cases. The training and testing results of thirty different BP-NN architectures using the sigmoid transfer function and the delta learning rule showed that the network with a single hidden layer having five hidden neurons is the best network with training and testing errors of 0.002 and 0.000 respectively. The evaluation results of the four models over a test samples indicated that the BP-NN outperformed the regression models with an average error of -2.76% and Mean Absolute Percent Error (MAPE) of 4.53%. The neural network model showed satisfactory performance therefore it can be used by both clients and contractors for estimating highway construction duration at the planning phase.
No. of Downloads :