•  
  •  
 

ORCID

Ahmed M. AbdelMouty: https://orcid.org/0000-0002-2734-9486

Keywords

Ensemble machine learning, Earthquake alert level classification system, Explainable AI, Attention mechanism

Article Type

Original Article

Abstract

Earthquakes are one of the most life-threatening natural events, and scientists are always concerned about their forecast. Standard machine learning (ML) algorithms have difficulty earthquake alert level classification (EALC) problem. The study presents an ADASEL framework for EALC. The data is merged using a stacked ensemble learning approach, including base learners and meta-models. The basic learners include Extra Tree, random forest (RF), lightGBM, XGBoost, CatBoost, and Gradient Boosting. The outputs are optimized using an attention-based multilayer perceptron meta-model (MLP), LSTM, and Transformer. The technique was tested using earthquake dataset. The data preprocessing steps are applied to enhance quality of the dataset. The ADASEL method outperforms baselines in terms of macro-F1, MCC, precision, and recall. The ablation study is conducted in this study to show different analysis of the proposed approach. Explainable Artificial Intelligence (XAI) is used with the proposed approach to show the highest importance features effect on the prediction task. The results show the ADASEL method with transformer is the best model with 94.84 accuracy better compared to different standard ML models. It can be seen from the experiments that the proposed attention-based adaptive stacking approach is robust and efficient. It has been proven by further analysis that the proposed method is valid and interpretable.

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Share

COinS