An Advanced Multi-Model Fusion Approach for Earthquake Prediction with Uncertainty Estimation from Seismic Sensor Streams

Abstract

The ability to predict when an earthquake will happen along with issuing an early warning is important in order to prevent problems from occurring due to seismic activity. The large amounts of non-linear and complex nature of seismic data usually leads to traditional machine learning models not having the ability to predict accurately or reliably. We improved the earthquake alert classification process through the development of the LGCET framework by a combination of the algorithms - LightGBM, CatBoost, and ExtraTrees. The use of the proposed LGCET framework was developed in order to take advantage of the strengths from both boosting as well as tree-based learning. LightGBM will efficiently handle large amounts of data and complex feature interactions. The features in the dataset are most likely represented as categorical features; this makes CatBoost the algorithm of choice for improving the handling of categorical features. The use of ExtraTrees has provided an increase in the robustness of the predictions made because their variances are reduced. The integration of advanced preprocessing and feature engineering methods (such as data balancing, feature encoding, and extracting relevant attributes of interest from a seismic event such as magnitude, depth, and spatial relationships) were included in the modelling process. The dataset that was used to train and evaluate the model came from Kaggle and USGS, thereby providing a diverse and reliable dataset. In addition to providing a predictive model, the system is designed with an automated alert function as well as visualisation capabilities to enhance user experience. The experimental results indicate that the LGCET framework will yield an accuracy rate of approximately 98% in comparison to traditional models such as SVC, KNN, or XGBoost. The results provide a reliable, scalable method for the classification of earthquake alert prediction; therefore, it will contribute to improved disaster preparedness and decision-making.