Development of a lean-based ergonomic index using machine learning to assess the health, safety, and performance of construction workers

This study presents the development of the construction labour health, safety, and performance index (CLHSPI)—a lean-based ergonomic assessment model aimed at optimizing the well-being and productivity of construction labourers. Data were obtained from real-world construction sites and preprocessed with techniques such as imputation, outlier elimination (using Z-score and IQR), and normalization to assure data integrity. To forecast and validate the index, four machine learning methods were used: gradient boosting machines (GBM), support vector machines (SVM), chi-squared automatic interaction detection (CHAID), and artificial neural networks (ANN). These models were trained on the improved dataset with feature selection techniques such as principal component analysis (PCA) and recursive feature elimination (RFE). In terms of predicted accuracy and generalizability, GBM outperformed the other models. The CLHSPI is a reliable and scalable technique for identifying high-risk labour circumstances, assisting decision-makers in improving site safety, working environments, and employee happiness.