Customer　transaction　fraud　detection　is　an　important　application　for　both　the　public　and　banks　and　it　is　becoming　a　heated　topic　in　research　and　industries.　Many　data　mining　techniques　have　been　utilized　in　financial　sys-tem　to　save　consumers　millions　of　dollars　per　year.　In　this　study,　we　presented　a　Xgboost-based　transaction　fraud　detection　model　with　some　feature　engineering　and　visualization.　The　dataset　is　from　IEEECIS　Fraud　Detection　Competition　on　Kaggle,　which　is　a　well-informed　data　science　organization.　The　study　indicated　that　xgboost　based　model　outperformed　the　other　three　methods　including　Support　Vector　Machine,　Random　Forest　and　Logistic　Regression.　As　to　two　feature　selection　methods,　Xgboost　performed　better.　Our　best　model　achieved　95.2%　roc　auc　score　on　leader-board　and　defeated　other　98　percent　participants.　©　2020　IEEE.