The construction industry faces elevated safety risks due to complex operational environments and diverse project demands, particularly during the structural work of apartment construction projects. Although regulations mandate minimum safety and health management expenses (SHME), current standards inadequately reflect the unique hazards associated with structural work. Traditional estimation methods, such as regression analyses and questionnaire surveys, fail to capture the nonlinear interactions among key cost variables, thereby limiting prediction accuracy and interpretability. This study proposes an accurate and interpretable prediction framework to address these limitations.
Actual data from 30 completed apartment structural work projects were analyzed. Five filter-based feature selection methods – including mutual information, analysis of variance F-test, minimum redundancy maximum relevance (mRMR), Pearson correlation, and grey relational analysis – were employed to identify key predictors. Five machine learning models – support vector regression, random forest regression (RFR), extreme gradient boosting, light gradient boosting machine, and artificial neural networks – were trained and validated using five-fold cross-validation. SHapley Additive Explanations (SHAP) were used to quantify the contribution of each variable.
The mRMR-based RFR model delivered the best predictive performance on the independent test set, achieving a mean absolute percentage error of 0.08. SHAP analysis explicitly identified material costs and safety-related staffing as critical factors significantly influencing SHME.
This study uniquely integrates multiple filter-based feature selection methods, ensemble machine learning algorithms, and SHAP interpretability for predicting SHME in apartment structural work. The proposed framework offers transparent decision support, assisting governments and construction firms in safety budgeting, resource allocation, and the enhancement of safety management practices.
