Aim: Cardiac arrest is a common, acute, and critical disease in clinical practice. The electroencephalogram (EEG) signal of patients with cardiac arrest can objectively predict their nerve recovery level. Compared to using only a random forest classifier, we use multiple feature selection algorithms and deep bagging aggregation of multi -classifier to improve the accuracy of forecasting. Methods: We adopted ReliefF and Principal Component Analysis(PCA) algorithms for filtering feature selection on the data to obtain different subsets of data, and then performed multi-classifier fusion using Deep Bagging aggregation on different subsets of data. The ReliefF algorithm gives different weights according to the correlation between various features and categories and removes the characteristics lower than the threshold, and the PCA algorithm reduces the dimension of high-dimensional data. In Deep Bagging, we use the data subset we have obtained before to train multiple models. In the testing stage, we will average or vote on the prediction results of multiple models to get the final prediction results. Results: The preliminary classifiers of our team AHU lab have achieved challenge measurement scores of 0.09, 0.22, 0.10, and 0.34 on the hidden verification set. The outcome of AUROC and AUPRC are 0.65 and 0.76 respectively. Conclusions: We provide a practical solution to the neurological prediction of patients with cardiac arrest and specific recovery level suggestions from EEG. Different from a single model prediction, we have used a variety of feature selection methods and Deep Learning integration to make the prediction process more reliable and efficient.