<![CDATA[Earthquakes are one of the most uncertain geological disasters and have the potential to cause significant damage in North Sumatra. The pattern of earthquake occurrences is influenced by complex tectonic conditions, which are affected by changes in seismic activity over time and the characteristics of the location. The aim of this research is to develop a spatiotemporal-based AutoML earthquake vulnerability classification model optimised with Bayesian optimisation methods. The seismic data used includes epicentre coordinates, depth, magnitude, time, frequency, and distance from active earthquake sources. Data preprocessing, spatial and temporal pattern analysis, feature engineering, vulnerability class determination, training several classification algorithms through an automated machine training framework, and hyperparameter optimisation using Bayesian optimisation. To assess the model's performance, accuracy, precision, recall, F1-score, area under the curve, and confusion matrix metrics are used. To reduce the possibility of bias and ensure that the model can be generalised to various locations and periods of occurrence, spatial and temporal validation are used. It is expected that the research results will produce a classification model with higher accuracy and stability than conventional classification methods. Next, the best model is used to divide the area into low, medium, and high vulnerability categories. It is hoped that this research will help develop a more adaptive, objective, and efficient data-based earthquake vulnerability mapping system that will assist in the decision-making process and disaster mitigation planning in the North Sumatra region.]]>
Copyrights © 2026
