In　view　of　the　unbalanced　data　set,　the　original　intrusion　detection　method　cannot　accurately　detect　the　network　security　intrusion　behavior.　An　improved　smote　and　XGBoost　intrusion　detection　method　is　proposed.　In　terms　of　the　diversity　of　security　data　types,　the　data　is　processed　by　the　use　of　one-hot　encoding.　In　terms　of　data　imbalance,　using　the　improved　smote　technique　to　oversample　for　the　class　with　a　small　amount　of　data　and　downsample　for　a　class　with　a　large　amount　of　data.　Finally,　using　the　KDD99　security　data　set　to　conduct　experiments　and　calculate　the　missing　rate,　average　precision,　average　recall　rate　and　F-Score　of　the　model,　and　compare　with　the　original　training　method　results.　The　results　show　that　the　method　has　high　accuracy　and　low　missing　rate　in　intrusion　detection.