Author(s)

Jingchuan Xu, Yuanyuan Ren, Hao Zhang, Shuyan Fu

Corresponding Author

Jingchuan Xu

Abstract

This paper presents a flood occurrence prediction method based on the entropy weight method and TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) combined with ensemble machine learning. Initially, the flood probability is clustered into three risk levels—low, medium, and high—using the K-means clustering algorithm, and the minimum occurrence probability, maximum occurrence probability, average flood probability, and proportion quantity for each level are calculated. Subsequently, feature extraction of risk indicators is performed through R-type clustering, and similarity measures such as the Pearson correlation coefficient and cosine similarity are utilized to assess the degree of similarity between variables, followed by variable clustering analysis. Furthermore, an early warning evaluation model based on the entropy weight TOPSIS is established, where information entropy is used to quantify indicator weights, reflecting the importance of indicators in the decision-making process. The model calculates information entropy, indicator weights, standardized decision matrix, positive and negative ideal solutions, and comprehensive evaluation index to rank the sample risk levels. Additionally, this paper explores the sensitivity analysis of the model by perturbing the weights of each feature indicator and the flood characteristics, verifying the model's robustness under minor perturbations. Finally, a flood probability prediction model based on different base classifiers is constructed using the Stacking ensemble learning algorithm, and cross-validation methods are employed to enhance the model's generalization capability. The experimental results demonstrate that the Stacking model performs best on evaluation metrics such as Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Coefficient of Determination (R²), and Maximum Absolute Error (MaxAE), proving its superiority in predicting the probability of flood occurrence.

Keywords

Flood Prediction; Entropy Weight Method; TOPSIS; Ensemble Machine Learning; Stacking Algorithm