interaction2

interaction

plot7

This project developed and evaluated a series of supervised machine learning models for tsunami event prediction based on earthquake-related features. By systematically training, tuning, and validating multiple algorithms—including Random Forest (initial and tuned), Logistic Regression, Support Vector Machine (SVM), Decision Tree, and K-Nearest Neighbors (KNN)—the study aimed to identify an optimal predictive framework balancing accuracy, recall, and interpretability. Across all models, the tuned Random Forest emerged as the most robust and balanced performer, achieving high accuracy, precision, recall, and AUC, thus demonstrating strong discriminative capacity and generalization potential. Its ability to correctly identify nearly all tsunami-generating earthquakes while minimizing false alarms highlights its suitability for real-world early warning applications. The Decision Tree model also performed admirably, showing perfect recall, indicating complete sensitivity to tsunami occurrences—though at the cost of slightly higher false positives. The SVM demonstrated consistently high classification accuracy but exhibited minor imbalance between precision and recall, implying limited reliability in uncertain boundary cases. Logistic Regression and KNN, while conceptually simpler, showed reduced predictive strength and sensitivity, particularly KNN, which struggled to detect true tsunami cases effectively. The integrated model comparison confirmed that ensemble-based approaches (particularly the tuned Random Forest) significantly outperformed linear and instance-based methods, suggesting that complex, nonlinear interactions among seismic variables are key to accurate tsunami risk modeling. Visual analyses, including ROC curves and actual-versus-predicted charts, further substantiated the Random Forest’s dominance in maintaining high sensitivity and specificity simultaneously. In summary, this project established a reproducible, data-driven workflow for tsunami occurrence prediction. It demonstrated that ensemble learning, when properly tuned and evaluated using comprehensive metrics, can serve as a reliable foundation for automated tsunami early warning systems. These findings not only contribute to improved disaster preparedness and mitigation strategies but also offer a methodological reference for future geohazard classification research

needed

first cycle

manova

Lab 1

MANOVA