Optimization and performance evaluation of machine learning classifiers for predicting construction quality and schedule

In recent decades, numerous predictive machine learning (ML) models have been developed within the field of construction management. However, comprehensive evaluations of supervised, data-driven methods that can address the complexity inherent in construction project data remain limited. To bridge this gap, this paper utilized a large-scale, publicly available dataset from the Public Construction Intelligence Cloud (PCIC), comprising 1015 projects characterized by 499 distinct defect categories. Nine supervised ML classifiers were evaluated on two distinct prediction tasks: (i) classifying construction quality into four categorical clusters, and (ii) predicting construction schedule status as either ahead or behind schedule. Each ML model underwent hyperparameter tuning during training to determine optimal parameter combinations, resulting in highly optimized predictive models. Among them, the Multilayer Perceptron (MLP) achieved the highest accuracy, 94.1 % (F1 score: 0.902) for quality prediction and 98.4 % (F1 score: 0.984) for schedule prediction, demonstrating its effectiveness in construction data analysis.