Comprehensive Analysis of Flood Risk Using Regression Models and Interpretable Machine Learning Techniques

Since flooding incidences are becoming more frequent and serious globally due to climate change, flood prediction has become more crucial than ever. Recently, in forecasting systems, Machine Learning (ML) is being utilized to build efficient and accurate systems. In this work, ML algorithms Linear, Ridge, Lasso, Polynomial, Support Vector, Decision Tree, Random Forest, Gradient Boosting, and Bayesian Ridge Regression (BRR) are trained using publicly available flood prediction dataset and investigated for their effectiveness. The performance of each model is measured using MSE, RMSE, MAE, and R-squared (R2). Out of all the models, BRR shows the best performance with an R2 value of 1.00. Using the techniques of Explainable AI, namely LIME and SHAP, we further elucidate the model to help obtain greater transparency and reliability of results in predictions