The Construction Safety Risk Assessment Method for Hybrid Integrated Learning Based on SNS

ABSTRACT

This study addresses the limitations of existing machine learning and deep learning methods in construction safety risk assessment, particularly their limited generalization across diverse contexts and their difficulty in handling complex, high‐dimensional data. A hybrid ensemble learning framework that integrates the social network search (SNS) algorithm and t‐distributed stochastic neighbor embedding (t‐SNE) technology is proposed to enhance prediction accuracy and computational efficiency. A comprehensive, multi‐dimensional risk assessment database was developed through questionnaire surveys and on‐site data collection. t‐SNE was employed to reduce data dimensionality and extract key features, which were then used to train a hybrid ensemble model combining extreme gradient boosting (XGBoost), a backpropagation neural network (BPNN), random forest (RF), and CatBoost. The SNS algorithm was applied to optimize model parameters, further improving predictive performance. Results indicate that the proposed approach achieves high accuracy and strong generalization capability, outperforming traditional single‐model methods. This framework offers a practical and effective tool for supporting construction safety management, enabling proactive risk prevention, and contributing to safer construction environments.