An Ensemble Machine Learning Approach for Predicting Flood Based on Meteorological and Topographical Features: A Comparative Study in Kalu Ganga River Basin, Sri Lanka

Floods are one of the most frequently happening disasters in Sri Lanka that causes severe damage in terms of loss of lives and property damage. Kalu Ganga is one of the river basins, most prone to floods in Sri Lanka. It is stated from studies, that machine learning approaches produce higher accuracy, and can be developed faster and more cost-effective than conventional methods of flood prediction. This study aims to enhance the accuracy of flood prediction in the Kalu Ganga river basins in Sri Lanka using ensemble of methods. The study focuses on six catchment areas such as Kalawana, Ayagama, Kuruwita, Pelmadulla, Elapatha and Kahawatta. The features considered are based on meteorological and topographical aspects. The methodology involves collecting and preprocessing the data followed by feature selection and developing predictive models. The accuracies of the models are evaluated using F1-score. The F1-score, a widely recognized measure of a model's accuracy, balances precision and recall. Specifically, it considers both false positives and false negatives, offering a nuanced evaluation of the model's performance. In the context of flood prediction, where the consequences of both false positives (incorrectly predicting a flood) and false negatives (failure to predict an actual flood) are severe, the F1-score proves to be a relevant and insightful metric. The results show that the Bagging classifier with decision tree as the estimator followed by the use of wrapper method backward feature selection and preprocessing has high F1-score in flood prediction of the Kalu Ganga river basin. This study demonstrates that ensemble of methods can effectively enhance the accuracy of flood prediction in the Kalu Ganga river basin.