Smart Disaster Prediction: A Unified Machine Learning Approach for Earthquakes and Floods

Abstract Natural disaster prediction is vital for the minimization of effects of these calamities on human life and infrastructure. The aim of this research work is to find an application of machine learning algorithms in earthquake magnitude and flood event prediction. This is achieved using historical seismic data, provided by the Indian Meteorological Department and United States Geological Survey. This paper was trained by optimizing the Random Forest model using RandomizedSearchCV. The model had a MAE of 14.5%, RMSE of 19.8%, and an R-squared value of 0.75, thus it proved to be efficient in the prediction of earthquake magnitudes. For flood prediction, a Voting Classifier ensemble model was used that comprised XGBoost, Random Forest, KNN and Decision Tree and, for the prediction of flood events with environmental and socioeconomic factors. The ensemble model presented significant results with an accuracy of 93.64%, precision of 89%, recall of 85%, and an F1-Score of 87%, which shows how this model is robust in flood-non-flood classification. Results therefore indicate the potentiality that machine learning models have for improving disaster prediction by indicating the strength of the ensemble methods in improving early preparedness, resource allocation, and early warning systems.