Estimating Tsunami Run-Up Triggered by Landslides Using Support Vector Machine

Accurate prediction of tsunami run-up heights induced by landslides is essential for coastal hazard assessment and early warning systems. This paper presents a machine learning approach based on Support Vector Machines (SVM) to estimate maximum run-up values using synthetic time series data. The raw sequences were transformed into nine statistical features through feature extraction, and a Relief-F-based feature selection strategy was applied to rank their importance. Successive models were constructed by iteratively eliminating the least relevant features, enabling performance evaluation under reduced input complexity. Among the nine models, the best results were achieved with Model-2, which used eight features and reached an RMSE of 0.36 m and an $\mathrm{R}^{2}$ of 0.997. Remarkably, Model-3, using only seven features, demonstrated comparable performance. These findings emphasize the effectiveness of combining feature engineering with SVM to develop interpretable and efficient predictive models. Compared to previous deep learning-based approaches, this framework delivers similar accuracy with significantly lower computational cost. These findings suggest that SVM can be a reliable alternative for tsunami run-up prediction, particularly in scenarios where observational data are scarce or difficult to obtain.