Classification of Flood Disaster Risks with the Use of Gradient Boosting Algorithm

Abstract

This study used base and ensemble approaches to classify the flood disaster risks in a local provincial capital in the Philippines using an intelligent methodology based on machine learning. It focused on Gradient Boosting Algorithm with Decision Trees as base classifiers/estimators. The researchers consulted with experts to determine the weights of causative factors to fluvial flooding, which were then classified into four (4) risk levels using the Quantile Method and the Exponential Regression for missing value imputation. The K-fold cross-validation was used to validate the proposed algorithms. The experiment shows that Gradient Boosting Algorithm is the most appropriate model for the disaster data with the score of 80.00%, more than 70% in all the classification criteria (accuracy, precision, recall f1-score), respectively.