Optimized Tsunami Vulnerability Area Classification using Hybrid Fuzzy-SVM Model Through Spatial Data Processing

Abstract

Tsunami disasters pose serious threats to human life and coastal infrastructure and require accurate mapping of tsunami-prone areas for effective disaster mitigation and coastal planning. Machine learning methods, including weighted overlay and Support Vector Machine (SVM), are widely applied but often face limitations in identifying transitional vulnerability zones with gradual class boundaries. This study proposes a hybrid fuzzy-SVM approach to improve the accuracy of tsunami vulnerability classification, which is an important parameter in classifying tsunami vulnerability. Three geospatial parameters elevation, land cover, and inundation extent, were used as primary inputs, each transformed through fuzzy membership functions to handle uncertainty and spatial ambiguity. The Fuzzy Vulnerability Index (FVI) smooths the resulting values ​​and then classifies using SVM with linear and RBF kernels under a one-vs-rest scheme to generate multi-class vulnerability maps across the southern coast of East Java. Experimental results demonstrated that the proposed hybrid fuzzy–SVM outperformed both conventional SVM and weighted overlay methods. The model achieved an overall accuracy of 91.3%, precision of 0.911, recall of 0.910, and F1-score of 0.910, indicating strong agreement between predicted and reference vulnerability maps. Overall, the hybrid fuzzy–SVM framework provides a more flexible and data-driven approach to tsunami vulnerability assessment.