Improved V-detector algorithm based on bagging for earthquake prediction with faults

With the highly nonlinear relationship between various seismic feature indicators and earthquakes, the researchers can hardly construct an earthquake model. Meanwhile, the lack of samples for destructive earthquakes also leads to inaccurate medium-to-short-term earthquake magnitude predictions. Therefore, this study proposes a novel model for earthquake prediction, named variant detector bagging algorithm (V-detector-bagging). First, we adopt the geological distribution of earthquakes and faults as a criterion to select the appropriate history catalog data area, and the seismic indicators are calculated through the Gutenberg–Richter laws and Panakkat indicators. Then, we propose the V-detector-bagging algorithm, which combines the V-detector algorithm with the bagging method. The proposed algorithm converts the self-tolerance process into a cycle process, during this process, different selves guide samples to spread widely, generate various detectors covering more wide nonself areas, fill holes that are not covered by detectors in nonself areas, and reduce the false negative rate. Thus, the V-detector-bagging algorithm improves the detection performance of the V-detector. Finally, through experimental validation analysis, the proposed algorithm ranked first in the detection rate on Sichuan and Xinjiang catalog data compared to the popular machine learning methods used in predicting earthquakes and the original V-detector algorithm, with Xinjiang yielding the best results.