Risk Assessment and Prediction Based on Generalized Adaptive Backpropagation Neural Networks

This paper introduces a novel risk evaluation and prediction model based on the GABP neural network. Initially, a K-means clustering model categorizes events into three risk levels: high, medium, and low. Subsequently, a TOPSIS entropy weight model is developed to select various risk events and establish corresponding early-warning evaluation models. The model’s effectiveness is validated through sensitivity analysis using a confusion matrix, achieving a sensitivity value of 0.938, and F1 scores above 92% across training, testing, and cross-validation datasets, with the training set reaching 93.6%. Further, the paper explores regression models including Random Forest and XGBoost, concluding that the GABP neural network optimized with a genetic algorithm exhibits superior performance, with an optimal evaluation effect at K=5 and a testing accuracy of 95.87%. These findings underline the model’s robustness and high predictive accuracy in risk assessment.