Predicting earthquake damage using a PAR-CGA: parallel cost-sensitive genetic algorithm- based rule classifier

Celik M. E., Özmen M., Şahin Ö.

COMPUTATIONAL GEOSCIENCES MODELING, SIMULATION AND DATA ANALYSIS, cilt.29, sa.6, ss.29-48, 2025 (SCI-Expanded, Scopus)

Özet

Earthquakes are natural disasters that cause great destruction and loss of life. The most critical thing to do after an earthquake is to determine the damage status of the structures quickly and reliably. Earthquake damage datasets are multi-class, high-dimensional, big, and imbalanced distributed. Such datasets face a serious classification challenge: running time, classification performance. To address these issues, this study employs PAR-CGA: Parallel Cost-Sensitive Genetic Algorithm-based Rule Classifier. PAR-CGA uses vectorization and GPU (Graphics Processing Unit) parallelization techniques to speed up the algorithm. This study uses PAR-CGA to predict the damage to buildings affected by the Samos and Sivrice earthquakes. Besides, the algorithm's effectiveness was tested on publicly available datasets similar to the earthquake dataset and compared with well-known rule classifier algorithms. Statistical analysis and experiments show that PAR-CGA outperforms other algorithms regarding Macro F1 values.

Earthquakes are natural disasters that cause great destruction and loss of life. The most critical thing to do after an earthquake is to determine the damage status of the structures quickly and reliably. Earthquake damage datasets are multi-class, high-dimensional, big, and imbalanced distributed. Such datasets face a serious classification challenge: running time, classification performance. To address these issues, this study employs PAR-CGA: Parallel Cost-Sensitive Genetic Algorithm-based Rule Classifier. PAR-CGA uses vectorization and GPU (Graphics Processing Unit) parallelization techniques to speed up the algorithm. This study uses PAR-CGA to predict the damage to buildings affected by the Samos and Sivrice earthquakes. Besides, the algorithm's effectiveness was tested on publicly available datasets similar to the earthquake dataset and compared with well-known rule classifier algorithms. Statistical analysis and experiments show that PAR-CGA outperforms other algorithms regarding Macro F1 values.