Prediction of Cost Overruns in Solar EPC Projects Using Machine Learning Techniques: A Data-Driven Study in India

Abstract

One of the challenges that continuously occur in the context of an Engineering, Procurement, and Construction (EPC) project, especially in the fast-developing solar energy industry in India, is cost overruns. The conventional cost estimation methods are usually not effective in capturing the multifaceted and dynamic nature of relationships among the project variables; hence, leading to huge discrepancies between the estimated and actual costs. The proposed study will propose a machine learning-based framework based on data analysis to forecast cost overruns on solar EPC projects in India.

Some of the important influencing factors considered in the research are project size, labor cost, material cost, project delays, location characteristics and environmental conditions. The analysis of a structured dataset of solar EPC projects is performed with the help of multiple machine learning models, i.e., Linear Regression, Support Vector Machine (SVM), Random Forest (RF), and Artificial Neural Networks (ANN). Mean Absolute Error (MAE), Root Mean square error (RMSE), and coefficient of determination (R 2 ) are used to evaluate model performance.

The findings suggest that ensemble methods, specifically, Random Forest, are more accurate in prediction than the conventional statistical models. The research makes some contribution to the body of literature because it narrows down to specifically analyzing Indian solar EPC projects and incorporating technical, financial, and environmental variables into predictive modeling. The results have practical implications on the project managers and the policy makers to improve the accuracy of cost estimation, minimize financial risks and make better decisions in renewable energy infrastructure projects.