A Hybrid Mamdani Fuzzy Inference System and Generalized Regression Neural Network for Cost and Time Overrun Prediction in Expressway Construction Projects

Abstract

In construction projects like expressways, risks and uncertainties are unavoidable and have the potential to significantly alter the expected outcome, which would be detrimental to the project's success. Risk is one of the main causes of productivity and efficiency losses in the construction sector that results in project demise, disputes, costs, and time overruns. The failure to complete the construction project within the stipulated time and estimated cost due to various risk factors is a major problem nowadays. This work analyzes the risk factors resulting in cost and time overrun in expressways construction projects using the Mamdani Fuzzy Inference System and Generalized Regression Neural Network(H-MFIS-GRN2) hybridization. Initially, the MFIS is used to find and measure potential dangers in the building process by analysing expert-made fuzzy rules. Defuzzification of the outputs yields clear risk severity values, which are further weighted with the use of mean scores and the Relative Importance Index (RII). Learning to anticipate budget and schedule overruns, MFIS doesn't provide the final product risk variables as normalized and prioritized inputs. The GRN2 model trains the observed risk factors with a pattern using the Gaussian activation function in a single pass. The model was validated using data from 27 areas from completed highway building projects. The trials show that the H-MFIS-GRN2 model outperforms baseline models. These baseline models are HRF-GA, H-AHP-ANN, and F-MRA. The H-MFIS-GRN2 model has 92.5% accuracy and 5.3% MAPE. Comparison analysis has increased forecast accuracy and interpretability, helping prioritize and minimize key risk variables. Fuzzy logic and neural networks can be used together to detect early risk in major infrastructure projects due to their strengths in uncertainty and learning.