Epilepsy　is　a　common　neurological　disease,　incidence　of　a　disease　is　higher,　it　is　mainly　caused　by　excessive　discharge　of　brain　neurons,　attack,　the　characteristics　of　transient,　repeatability　and　consistency,　in　recent　years　the　rise　in　the　number　of　global　disease,　epilepsy,　serious　damage　to　human　disease,　so　the　research　of　epileptic　seizure　prediction　is　very　necessary.In　the　traditional　epilepsy　prediction　framework,　due　to　the　traditional　convolutional　neural　network　(CNN)　network　model　and　long　and　short-term　memory　network　LSTM　network　hidden　layer　units　and　learning　rate　will　be　set　up　improper　overfit　and　instability,　will　lead　to　poor　network　prediction　effect.　Therefore,　the　combination　with　CNN　network　and　LSTM　network　is　selected　as　the　classification　model　in　seizure　prediction　to　form　the　epilepsy　prediction　model.　The　main　research　content　of　this　paper　are　as　follows:　before　extracting　the　seizure　EEG　data,　using　wavelet　transform　to　preprocess　EEG　signals,　and　select　sample　entropy　as　a　feature　extraction　method　to　predict　the　processed　data　using　CNN-LSTM,　the　prediction　accuracy　reached　87.8%,　4.3%　over　CNN　and　5.0%　over　LSTM.It　shows　that　the　present　method　can　solve　the　classification　problem　of　existing　epilepsy　datasets　well.　©　2022　IEEE.