Objective: Electroencephalogram (EEG) monitoring is a powerful tool for neurological prognosis after cardiac arrest (CA), removing the subjectivity of the physician's judgment. However, prior approaches largely rely on snapshots of the EEG, without taking advantage of temporal information. We attempt to use a multivariate time-series deep neural network (DNN) to automate the analysis of continuous EEG data. Methods: Our team, UCASFighters, proposed a DNN combining long- and short-term time-series network (LSTNet) and random forest (RF) classifiers, using data from a multicenter cohort of 607 cardiac arrest patients. LSTNet used the convolution neural network and recurrent neural network to extract temporal dynamics from EEG and predict long-term neurological outcomes, while the RF used several key time-domain and frequency-domain features of EEG data and demographic information. In addition, attention mechanism was added to increase the weight of EEG at specific periods of time to focus attention on more useful information. The outputs of LSTNet and RF were fused to obtain the final prediction probabilities of neurological outcomes. Model performance is evaluated using 5-fold cross validation. Results: Model performance increased with EEG duration, with area under the receiver operating characteristic curve increasing from 0.64 (95% CI 0.61-0.68) at 12h to 0.71(95% CI 0.68-0.74) at 72h. Sensitivity of good and poor outcome prediction was 65% and 63% at a specificity of 95%, respectively. Predicted probability was well matched to the observation frequency of poor outcomes, with a calibration error of 0.16 [0.14-0.19]. Conclusions: Our proposed method achieved the Challenge metric scores of 0.06, 0.22, 0.25, 0.42 on the hidden validation set. Our approach improves the prediction of neurologic outcomes for patients after CA and incorporating EEG evolution over time can substantially improve our model's performance. Our study demonstrates the value of using temporal information from continuous EEG data to predict neurological outcomes.