Aditya Gupta, Babita Sharma, Prasanta Chingtham

Earthquakes have posed significant hazards to human lives and infrastructure for as far back as can be recalled. This paper presents a Machine Learning (ML) based approach for earthquake magnitude forecasting spatially using the earthquake clustering in five selected zones of NW Indian region. Previous research efforts have primarily relied on empirical relationships and statistical models, which often struggled to capture the complex dynamics associated with earthquakes. However, with the emergence of ML techniques, the ability to analyze large datasets and uncover hidden patterns has significantly improved. We propose ML models to forecast earthquake magnitudes in the five identified earthquake zones present in NW Indian region using Random Forest and Support Vector techniques. For each earthquake, we utilize the latitude, longitude, depth, and zone information for model prediction. Our models obtain a cumulative weighted average (Root Mean Square Error) RMSE of 0.407 for the Random Forest Regressors and a cumulative weighted average RMSE of 0.420 for the Support Vector Regressors. Our results improve over previous results in the field due an emphasis on zone-based models. This study demonstrates the potential of machine learning techniques in earthquake magnitude forecasting which may be utilized for proactive measures in mitigating the impact of seismic events.