PREDICTION OF SEWAGE PIPELINE CONSTRUCTION DURATION BY INTRODUCING MACHINE LEARNING AND DEEP LEARNING APPROACHES

Abstract

Establishing project costs in construction is crucial for project success, typically done through regression methods for prediction. While these methods are common, novel regression methods are less practiced in construction management. This study explores both traditional and modern regression techniques, analyzing data from 83 sewage pipeline projects in South Korea. The study implemented state-of-the-art frameworks, including hyperparameter optimization and k-fold cross-validation, to evaluate statistic, machine learning and deep learning based regression models using R2 score, RMSE, MAE, and MSE. Results revealed that performance metrics don't always align with predictive accuracy. For instance, the random forest regressor achieved the best R2 score of 0.847 but ranked fifth in prediction accuracy. Moreover, polynomial regression outperformed novel methods with a 98.790% accuracy across the validation dataset.