Earthquake Magnitude Prediction: Comparative Study of Machine Learning and Deep Learning Methods

Earthquakes are natural disasters that cause significant damage to infrastructure and pose serious risks to human life, particularly in seismically active regions such as Indonesia. Accurate earthquake magnitude prediction is therefore an important component of early warning and disaster mitigation systems. This study investigates the use of machine learning and deep learning approaches for predicting earthquake magnitudes based on Indonesian earthquake data from 2000 to 2025, comprising 59,571 recorded events. Two experimental scenarios were explored models trained with all available features and models trained on reduced-dimensional data using Principal Component Analysis (PCA). The evaluated methods include Backpropagation, Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), Feedforward Neural Networks (FNN), Extreme Gradient Boosting (XGBoost), and hybrid combinations such as FNN + XGBoost. Model performance was assessed using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and R². Results show that boosting-based approaches, particularly XGBoost and FNN + XGBoost, consistently delivered the best performance, achieving the lowest MAE (0.1426) and highest R² (0.7534) when using the full feature set. PCA-based models exhibited a modest accuracy decline of around 8-10% but offered substantial computational gains, with reduced training time and memory usage. These findings highlight a trade-off between predictive accuracy and efficiency, suggesting that feature-rich models are more suitable for high-precision seismic risk analysis, while PCA-reduced models are more practical for real-time operational deployment in earthquake early warning systems.