This article explores flood disaster prediction using a systematic approach to data analysis and model development. Pearson correlation coefficients are computed and visualized through heatmaps to examine the relationship between indicators and flood probability, followed by regression analysis to understand influencing factors. The KMeans algorithm is used to classify flood risk into low, medium, and high levels, and feature importance is determined with random forest classifiers to assist in model construction. Various regression models—linear regression, decision tree, random forest, and gradient boosting—are trained and tested, optimized using particle swarm optimization, and evaluated through mean squared error and coefficient of determination. Finally, the model’s predictions are validated against test data, showing an R2 of 0.86 and an MSE of 0.0003, with Kolmogorov-Smirnov and Anderson-Darling tests revealing significant deviations from a normal distribution.