A Multi-Step Feed-Forward Backpropagation Neural Network Model: Predicting Project Cost Overrun Rating Factors on Construction Labor Productivity

Abstract

Construction Labor Productivity is a way of measuring the work rate of unit labor. This is usually described as output per man-hour. The project is at risk from financial exposure as an effect of low construction labor productivity which is vulnerable to different factors that can be categorized into manpower, management, resources and site condition, and events not attributable to the contractor. A multi-step feed-forward backpropagation neural network model for predicting cost overrun rating based on construction labor productivity factors was developed employing MATLAB 2019a, which will be utilized as a supplementary tool in construction management. The predictive model underwent a parametric analysis that determined the relationship of each subset upon having varying values. The variable importance of each factor and subset to the cost overrun was determined through Feature Reduction Approach. The predictive model is a financial indicator projecting the probable cost overrun rating of a project with the presence of the factors affecting negatively the construction labor productivity. With the utilization of the developed predictive model, the project management team will have better insights on the possible financial risks based on the present factors and alleviate its probable negative impact through the application of mitigations which is beyond proactivity.