Using Machine Learning Applications to Develop Models For Predicting The Impact Of Construction Claims On The Cost And Time Of Projects In Syria

Abstract

Most construction projects in Syria suffer from numerous claims due to various reasons, such as change orders, design errors, and other factors that negatively impact project time and cost. Therefore, this article aims to develop models capable of predicting cost and time increases resulting from these claims. The objective was achieved through a questionnaire distributed to engineers working in the public sector in both supervision and execution asking them about the impact of each claim on project cost and time. Then they were used to develop two separate cost and time prediction models. Three machine learning methods were selected for model development: artificial neural networks, gradient boosting, and random forests. After training the models on 71 projects, the model with the lowest error rate was selected. The analysis results showed that the random forest model performed best for cost prediction, achieving the highest coefficient of determination (R²) at 0.9940 and the lowest (RMSE) at 0.0922. For time prediction, the gradient boosting model, achieved the highest (R²) of 0.9607 and the lowest RMSE of 36.29 days. A tool utilizing the best models was developed and applied to a project to validate the methodology.