In　the　information　age,　people　get　more　and　more　data　from　various　channels,　especially　medical　data　and　business.　Data,　education　data　and　environmental　data.　In　the　field　of　data　mining　and　machine　learning　research,　decision　tree　has　great　advantages　in　knowledge　induction.　Decision　tree　constructs　tree　structure　to　obtain　classification　rules.　It　has　become　an　important　method　of　data　mining,　and　its　effective　mining　rules　have　theoretical　and　practical　significance.　Based　on　the　above　reasons,　aiming　at　the　problem　that　in　the　process　of　data　classification　prediction,　the　data　dimension　collected　in　reality　is　high,　and　there　are　redundant　attributes　that　lead　to　the　deviation　of　decision　tree　classification　results,　a　classification　prediction　method　based　on　classical　decision　tree　C4.5　algorithm　and　CART　algorithm　is　proposed.　The　main　research　work　is　as　follows:(1)In　this　paper,　the　data　set　is　pre-cleaned　to　screen　out　the　variables　that　are　meaningful　for　the　classification　prediction　experiment　and　improve　the　evaluation　efficiency.　Then　analyze　the　data　distribution　characteristics　of　the　variables　with　missing　values,　and　use　appropriate　filling　methods　to　fill　the　missing　data.(2)The　XGBoost　algorithm　is　used　to　evaluate　the　importance　of　features,　and　15　significant　features　related　to　classification　prediction　are　selected　to　improve　the　generalization　performance　and　training　efficiency　of　the　model.(3)Modeling　and　training　based　on　classical　decision　tree　C4.5　algorithm　and　CART　algorithm　to　obtain　the　classification　performance　of　the　model.　©　2023　IEEE.