Every Earth scientist would tell you predicting earthquakes remains one of the most difficult tasks, especially in complicated geological locations such as Sulawesi, which has multiple active faults. This would be the main reason for focusing the research on creating a novel, dependable method for predicting strong earthquakes of different magnitudes. A customized Combust, XGBoost, and LightGBM along with proprietary stacking method for predicting the different sizes of earthquakes is one of the main procedural techniques. The machine learning model utilized for predicting earthquakes was prepared with a historical dataset of earthquakes recorded by the United States Geological Survey (USGS) from (Mw 4.0-9.0) 1900-2024. Preparation was intricate as it demanded time format conversions, interval calculations, and standard feature scales. The model was able to predict the five main attributes: size, depth, location (latitude and longitude), and time between events with deep precision. The correlation and fit of the model were tuned incredibly with over 0.99 for R Squared and for the system’s accuracy with RMSE and MAE below 0.1. Moreover, the resulting images strongly correspond to the known locations of active faults in Sulawesi, confirming the model accuracy and reliability in geoscience. By providing more precise estimates, this research significantly contributes to Sustainable Development Goal (SDG) 11 (Sustainable Cities and Communities), which is crucial for reducing disaster risk and building stronger communities in earthquake-prone areas.
Madlazim et al. (2026) studied this question.