ANALYSIS OF MACHINE LEARNING ALGORITHMS FOR SEISMIC ANOMALY DETECTION IN INDONESIA: UNVEILING PATTERNS IN THE PACIFIC RING OF FIRE

Abstract

This study presents an integrated analysis of machine learning algorithms for the detection of seismic anomalies in Indonesia, a region within the volatile Pacific Ring of Fire. Employing Local Outlier Factor, Isolation Forest, and Elliptic Envelope algorithms, we processed a comprehensive dataset of seismic events characterized by latitude, longitude, depth, and magnitude. Our methodology involved standardizing these features and aggregating model predictions to establish a consensus mechanism for outlier detection. The results indicated that the vast majority of seismic events are consistent with the expected geological patterns, with a negligible percentage exhibiting anomalous behavior across the models. Through statistical analysis and visual mapping, we discerned that while anomalies are varied, they may correlate with specific seismic event features such as higher magnitudes or unique geographic locations. The consensus approach revealed a high-confidence subset of outliers, offering a focused direction for further seismological scrutiny. The study's implications extend to enhancing seismic risk assessment and early warning systems, providing a methodological framework for identifying seismic events that deviate from normative patterns. By outlining a scalable approach for anomaly detection, this research contributes to the predictive analytics tools available for disaster risk management and emergency preparedness, aiming to mitigate the impact of seismic hazards in seismically active regions

Keywords

• earthquake, unsupervised machine learning, anomaly detection, local outlier factor algorithm, isolation forest, elliptic envelope