Interpretable Machine Learning for Cost Estimation in Underground Pipe Installation

This study explores the application of interpretable machine learning (ML) models for accurately estimating excavation and installation costs associated with underground high-density polyethylene (HDPE) pipeline projects. Leveraging design data from seven projects across Asia, the research evaluates six regression models: linear regression, lasso regression, elastic net regression, random forest regression, XGBoost, and Gaussian process regression. Model performance was assessed using the R-squared (R2) metric, with XGBoost and Gaussian process regression achieving the highest R2 scores (above 0.89) on the testing data sets. Shapley additive explanations (SHAP) analysis reveals that pipe length, diameter, and depth are the most significant predictors, with pipe length contributing most substantially to cost estimations. The results indicate that ML models, particularly XGBoost and Gaussian process regression and stacked models, provide reliable cost predictions when diverse, high-quality data are available, offering a valuable tool for early-stage project budgeting and planning in the pipeline industry.