Research on Flood Risk Prediction Model Based on K-means and AHP-CRITIC Weighting Method

This study first used Pearson correlation analysis to examine the relationship between various indicators and flood occurrence, identifying the top ten most relevant indicators: “monsoon intensity,” “sedimentation,” “landslides,” “deforestation,” “agricultural practices,” “climate change,” “policy factors,” “population score,” “wetland loss,” and “urbanization.” Based on these indicators, a flood risk prediction model was constructed. The K-means algorithm was used to classify flood probabilities into three categories: “high risk,” “medium risk,” and “low risk.” The AHP-CRITIC method was applied to weight each indicator, and the weighted data was used to train a random forest model. The model’s hyperparameters were set to "n_estimators = 200," "random_state = 42," and "criterion = gini." Cross-validation with K-folds resulted in an average score of 0.531 and a standard deviation of 0.000434, indicating good robustness. Compared to existing models, the proposed model showed improvements in accuracy, precision, recall, and F1-score, and it demonstrated good generalizability through validation with data from multiple regions and climatic conditions. Sensitivity analysis revealed that indicators such as “monsoon intensity” and “landslides” had the greatest impact on the model’s predictions, providing important insights for model optimization.