Abstract: In recent years, Earthquake prediction has grown more crucial in recent years, spurred by the pressing need to reduce the ruinous impacts of natural disasters on populations and infrastructure. This paper presents a machine learning-based method that uses past seismic records to predict seismic events as one of three events: earthquake warning, explosion, or no earthquake. The categorization is done on the basis of prominent features like magnitude, depth, root mean square (RMS), and depth error. Various models were trained and validated, i.e., Random Forest, Decision Tree, and XGBoost, with the highest prediction performance shown by XGBoost. For instances labeled as earthquake warnings, a K-Means clustering algorithm is also used to identify the severity level—Minimum, Moderate, or Severe. In order to achieve interpretability and reliability of the model, LIME (Local Interpretable Model-Agnostic Explanations) is implemented and makes each prediction comprehensible, and the intuitive user-friendly web application developed with Flask enables end users to provide seismic parameters to generate real-time and transparent results. The framework went through robust unit, integration, and acceptance testing, establishing confidence in reliability as well as usability. Generally, this solution provides a strong and interpretable means of early earthquake detection that can aid in improved preparedness and response approaches.

Key Words: Earthquake Prediction, XGBoost, Random Forest, K-means Clustering, LIME (Local Interpretable Model-Agnostic Explanations), Flask Application, Severity Classification.

| DOI: 10.17148/IARJSET.2025.125196