Multi-Objective Optimized GAN-Bayes Model for Predicting Construction Accident Risk

Abstract

Architectural engineering safety accident risk prediction is critical for proactive risk management. Traditional models often suffer from insufficient prediction accuracy, hindering effective risk prevention. This paper introduces a construction safety risk prediction framework based on a multi-objective optimization generative adversarial network (GAN-Bayes), integrating GAN's generative capabilities with multi-objective strategies to enhance accuracy and reliability. Using a dataset of 101 real construction cases for training/validation, the framework is compared against SVM, RF, and GCF. Experimental results show significant improvements: the GAN-Bayes framework achieves 92.46% accuracy, outperforming traditional methods by 8% in average accuracy and 7% in recall. Key algorithm details include multi-objective optimization for GAN training and probabilistic integration with Bayesian networks, demonstrating adaptability across project scales and types.