Spatial Variability Analysis of Earthquake Magnitude and Depth in Indonesia using Random Forest based on Historical Data

Abstract

Indonesia lies at the convergence of three major tectonic plates, resulting in high seismic activity and complex spatial earthquake patterns. This study analyzes the spatial variability of earthquake magnitude and depth in Indonesia using a Random Forest Regressor based on historical earthquake data from the United States Geological Survey (USGS). A total of 24,981 earthquake records (1976–2026) were collected, and 21,886 valid observations were obtained after data cleaning. To reduce data heterogeneity caused by different magnitude measurement methods, the dataset was grouped into six magnitude types (Mb, Ms, Mw, Mwb, Mwc, and Mww), and separate Random Forest models were developed for each group using longitude and latitude as input features. The novelty of this study lies in the magnitude-type-specific modeling approach, which allows spatial patterns to be analyzed more consistently across different earthquake measurement scales. The models were evaluated using 10-fold cross-validation and standard regression metrics. Results show that spatial coordinates have limited ability to explain earthquake magnitude, as indicated by low R² values across all models. In contrast, depth prediction shows better performance, with the best result achieved by Mw (R² = 0.6801), indicating stronger spatial structure in earthquake depth distribution. Overall, the findings demonstrate that geographic location is more informative for modeling earthquake depth than magnitude. The proposed approach effectively captures spatial depth variability and provides a more structured analysis of earthquake characteristics across Indonesia.