Countless earthquake disasters occur every year in the world. In order to reduce earthquake disasters, earthquake early warning and forecasting technology is developing day by day. Recent advances in earthquake prediction have used machine-learned neural networks to build highly flexible point-process models to improve existing parametric models. This paper focuses on the characteristics of the earthquake process, uses causality and various machine learning and deep learning models to fit the earthquake process, judges the attributes of the earthquake, and establishes an earthquake prediction model. From the signal sequences detected by 170 sensors in total from eight earthquake events, it can be seen from the original signals that compared with artificial seismic waves, natural seismic waves have unstable fluctuations before their amplitude reaches the peak value, and the duration of the shock wave is long, and the peak value is relatively small. Low, relative fluctuations in power spectral density are obvious, and the signal correlation is low. Extract 99% of the variance features of the data through PCA (Principal Component Analysis), and then use Decision tree, GBDT (Gradient Boosting Decision Tree), RF (Random forest), and SVC (Support Vector Classification) to fit them again. The F1 score improved by 6.52 %.