The Debris Flow Risk Prediction Model Based on PCA-Elman

Accurate prediction of the risk levels of debris flows is crucial for devising effective disaster prevention and mitigation strategies. This study, based on debris flow sample data from Yunnan Province, initially employs Principal Component Analysis to reduce the dimensionality of the raw data, extracting key features and minimizing data dimensions. Subsequently, a 5-fold cross-validation method is utilized to segment the dataset into training and testing sets, and a predictive model integrating Principal Component Analysis with an Elman Neural Network (PCA-Elman) is constructed. The study investigates the impact of data imbalance and spatial variability on the model’s predictive accuracy and attempts to enhance the model’s generalization capabilities by employing the Adaptive Synthetic Sampling algorithm and incorporating samples from unknown regions. The results indicate that the model demonstrates high accuracy and generalization in predicting debris flow risks, with its Area Under Curve value, Average Precision value, and average precision scores surpassing those of traditional models, achieving an accuracy rate of 88.57%. After processing the data with the Adaptive Synthetic Sampling algorithm, the model’s accuracy rate increases to 98.33%. Furthermore, incorporating samples from unknown regions into the trained model significantly improves the prediction accuracy for debris flow risks in those areas. This research offers new insights into the assessment of debris flow hazards and disaster prevention and mitigation efforts, and provides a reference for the construction of predictive models for similar natural disasters.