WAVELET–CNN–LSTM FRAMEWORK FOR ACCURATE EARTHQUAKE MAGNITUDE PREDICTION FROM SEISMIC WAVEFORMS

Abstract

Earthquake magnitude is a fundamental seismic parameter for assessing the severity of earthquake intensity and potential hazard. Early and precise magnitude estimation plays a vital role in facilitating rapid decision-making for earthquake disaster mitigation and emergency response. However, due to the spectral nature of seismic signals, it is difficult to achieve reliable prediction of earthquake. Traditional machine learning techniques, including random forest and gradient boosting, relied on manually engineered features and less temporal modelling, which often resulting in suboptimal performance as well as suffer from limitations such as high computational complexity, increased processing time and overfitting. To overcome these challenges this research proposed an innovative regression model which captures long term temporal dependencies and multi resolution frequency data for earthquake magnitude prediction by decomposing the seismic signals into one dimensional discrete wavelet transform to extract multi scale coefficients which is processed through convolution layers to identify patterns. Subsequently, Long Short-Term Memory layers are employed to model long range dependencies, while fully connected layers conduct nonlinear regression to predict the normalized magnitude. The model is trained using the Adam optimizer with early stopping and mean squared error as the loss function. Training and validation have done on a station-based subset of the Standard Earthquake Dataset (STEAD) comprising 1500 events with an 80/20 train–validation split. The proposed framework achieves RMSE of 0.328, an MAE of 0.253, and an R² of 0.567, surpassing baseline models based on Random Forest and Gradient Boosting. These findings imply that combining wavelet-based multi scale representation with CNN–LSTM temporal learning facilitates efficient, accurate and station-robust earthquake magnitude predictions, which is suitable for scalable real-time seismic monitoring systems.