A Safety-Oriented Machine Learning Framework for TsunamiGenerating Earthquake Classification Using Calibrated Probabilities and Temporal Validation

Abstract—Rapidly determining whether a strong earthquake is capable of generating a tsunami is a time-critical problem for coastal warning centers. This paper presents an AI-driven binary classification framework that compares Random Forest (RF), Support Vector Machine (SVM), and Logistic Regression (LR) using seismic and geographic attributes available shortly after an event. The proposed pipeline includes data quality filtering, standardized preprocessing, class-weighted learning, probability calibration, temporal holdout validation, and threshold analysis oriented toward reducing missed tsunamigenic events. In the illustrative experiment reported in this draft, RF produced the strongest balance of accuracy, recall, F1-score, and area under the ROC curve, followed by SVM and LR. Magnitude and focal depth were the dominant predictors, while location and quality indicators contributed complementary information. The framework is intended as a rapid screening layer that can complement, not replace, physical source inversion and operational tsunami propagation models. Index Terms—Tsunami early warning, earthquake classification, Random Forest, Support Vector Machine, Logistic Regression, machine learning, seismic hazards.