Effective workforce forecasting is critical to strategic management in construction projects, particularly ensuring staffing is optimized for efficient and timely project completion. This aspect holds particular importance in densely populated urban environments, where the complexity and scale of high-rise building construction exacerbate the need for proactive maintenance strategies. This research aims to compare the efficacy of various machine learning models, including support vector machine (SVM), extreme gradient boosting (XGBoost), gradient-boosted decision trees (GBDT), and multi-layer perceptron (MLP) models, in predicting the workforce demand for building maintenance and repair work. Accordingly, 128 sets of data composed of seven indicators collected from the Hong Kong building maintenance and repair contractors were curated to serve as input-dependent variables to forecast the demands for different labor categories in addition to the overall workforce demands. The performance metrics for the four machine learning models were evaluated using mean absolute error (MAE), mean squared error (MSE), root mean squared error (RMSE), and determination coefficient (R2), with GridSearch employed to optimize their hyperparameters for enhanced predictive accuracy. The evaluation using MAE, MSE, RMSE, and R2 revealed that the XGBoost model achieved the most favorable results for predicting workforce demand on the testing sets (0.1790, 0.0582, 0.2412, 0.9915). Compared to MLP (0.1828, 0.0775, 0.2783, 0.8821) and GBDT (0.1828, 0.0828, 0.2878, 0.8739), XGBoost demonstrated a stronger performance in this specific task. Furthermore, the study also explored the potential of a long short-term memory (LSTM) model to forecast the gross value of building maintenance and repair work undertaken by main contractors, which strongly correlates with workforce demand. This approach aimed to capture the consistent growth trend observed across various maintenance activities. Consequently, the models developed in this study serve as valuable tools for optimizing workforce allocation and financial planning, particularly in dynamic and ever-changing environments.
Read full abstract