Feature Elimination and Comparative Assessment of Machine Learning Algorithms for Flood Susceptibility Mapping in Kerala, India

Floods are a catastrophic phenomenon with far-reaching consequences for infrastructure, the economy, and human lives, profoundly impacting regions globally. This study assesses flood susceptibility in four districts of Kerala: Ernakulam, Idukki, Kottayam, and Alappuzha. For the 2018 storm that caused flooding by Cyclone Ockhi, a flood map for the area was produced using Sentinel 1 satellite data in Google Earth Engine environment. The resulting map served as the foundation for further analysis. Based on the literature review, 16 potential flood causative factors were identified and incorporated into spatial maps in the Geographic Information System (GIS) environment. Analysis of the flood dataset was performed using Machine Learning (ML) algorithms, namely, Random Forest (RF), Decision Tree (DT), Gradient Boosting Machine (GBM), and XG Boost (XGB). Grid search was employed to identify the optimal hyperparameters for each algorithm, ensuring improved performance. Recursive Feature Elimination (RFE) was subsequently applied to select the most influential variables, resulting in a refined dataset. The chosen factors' feature importance scores were obtained, which were used to create the flood susceptibility map using the four ML models in a GIS environment. Evaluation metrics such as F1 score, accuracy, precision, recall, and ROC-AUC score were computed for each model, providing insights into the effectiveness of each algorithm in predicting the flood occurrence. The resulting flood susceptibility map for the best-performing ML model visually represents the varying levels of flood risk in the study area.