PREDICTING FLOOD RISK PROBABILITY USING ENVIRONMENTAL AND INFRASTRUCTURE-BASED INDICATORS A MACHINE LEARNING APPROACH

Abstract

Climate change, environmental degradation, infrastructure quality, and socioeconomic vulnerability collectively shape flood risk. This study developed machine-learning models to estimate flood probability using environmental, infrastructure, and socioeconomic indicators. The data analyzed were acquired from a database with 1,117,957 labelled observations, 20 predictor variables and the continuous outcome variable, Flood Probability. Data pre-processing involved checking for missing values, duplicate values, consistency of the data and elimination of the non-informative identifier variable. Six regression algorithms Decision Tree, Random Forest, Gradient Boosting, XGBoost, LightGBM, and CatBoost were evaluated using mean absolute error, mean squared error, root mean squared error, and coefficient of determination. CatBoost was the model that had the best predictive ability among the evaluated models, with a test MAE of 0.0220, an RMSE of 0.0267, and an R² of 0.7295. It had a good generalization and little overfitting because of its close similarity between training and testing. Dam quality, climate change, topography and drainage, siltation, monsoon intensity, deforestation and river management were found to be the top 7 predictors for the feature importance analysis. The results show that the flood probability is a result of the cumulative effect of environmental pressures, infrastructure condition and socioeconomic preparedness. The proposed framework used in conducting flood-risk screening, prioritizing infrastructure, disaster preparedness, and evidence-based planning. Real-time meteorological, remote-sensing, hydrological, and geospatial data should be included in future studies for temporal forecasting and external generalizability.