With　the　rapid　development　of　information　technology　and　the　informationization　reform　of　education　and　teaching　work,　students’relevant　information　is　recorded　in　various　forms.　It　is　of　great　significance　to　comprehensively　analyze　students’behavior　and　conduct　performance　prediction　and　warning　in　complex　educational　data.　Aiming　at　the　existing　problems　in　predicting　students’grades,　such　as　paying　less　attention　to　daily　behavior　data,　using　single　data　features,　coarse　data　granularity　and　insufficient　use　of　behavior　data,　a　grade　warning　model　based　on　ternary　deep　fusion　is　proposed.　It　starts　from　the　perspective　of　group　behavior　and　individual　behavior,　using　the　regular　feature　extraction　module　to　extract　regular　features　from　group　consumption　behavior　and　activity　behavior,　using　the　activity　feature　extraction　module　to　extract　active　features　from　individual　consumption　behavior,　and　using　the　diligence　feature　extraction　module　to　extract　diligent　features　from　individual　entry　and　exit　behavior　in　the　library.　Then,　the　three　types　of　features　are　fused　to　classify　students　and　provide　performance　warnings.　Experiments　on　public　datasets　have　shown　that　the　proposed　model　method　has　certain　performance　prediction　and　warning　capabilities.　©　2024　Journal　of　Computer　Engineering　and　Applications　Beijing　Co.,　Ltd.;　Science　Press.　All　rights　reserved.