Comparative analysis of AI models for cost prediction and feature importance in construction

Abstract

The construction industry has always been plagued by a high number of uncertainties; due to this, cost overruns are the most frequent challenge. To address this problem, this study aims to determine how the presence of risk increases the likelihood of cost overruns and to develop a predictive model that can help reduce them. The research focuses on designing and evaluating prediction models based on artificial intelligence (AI), trained on 101 data points. Performances were compared based on root mean square error (RMSE), coefficient of determination (R2) and mean absolute percentage error (MAPE). A user-friendly front-end tool was developed to increase the practicality of the study, enabling users to see instant cost overrun predictions. Feature importance for risk events has been conducted to find out the most contributing risk events for the prediction model. The adaptive neuro fuzzy inference system model outperformed the rest of the AI models with a testing RMSE of 3.4 and an R² of 0.688 and a validation RMSE of 0.35, an R² of 0.96 and an MAPE of 12.5%. Contract-related risk events were found to be most impactful for cost overrun predictions. This study showed a comparison between various types of AI models. It also developed a graphical user interface, bridging the gap between AI and industry applications. Also found out the important risk factor for cost overruns.