Pemodelan Random Forest Untuk Estimasi Magnitudo Momen Dari Tiga Detik Awal Gelombang P

Banten and West Java have high seismic activity, requiring rapid and reliable magnitudo estimation for earthquake early warning. This study developed a Random Forest model to estimate moment magnitudo (Mw) from the first three seconds of P-wave signals. BMKG strong-motion data from 2019–2025 underwent preprocessing and extraction of 16 amplitude, velocity, acceleration, energy, distance, and logarithmic features. Event ID was retained as a unique identifier for each earthquake in dataset preparation. Using GroupShuffleSplit, 4,209 samples were divided by Event ID into 3,394 training and 815 testing samples. Hyperparameters were optimized through Grid Search with five-fold GroupKFold validation based on Event ID. On the testing data, the model achieved an MAE of 0.159, an RMSE of 0.198, and an R² of 0.6115, indicating moderate predictive performance. Random Forest can support rapid initial magnitudo estimation, although broader data and improved features remain necessary.