It　is　very　meaningful　to　build　a　model　based　on　supply　chain　data　to　determine　whether　there　is　fraud　in　the　product　transaction　process.　It　can　help　merchants　in　the　supply　chain　avoid　fraud,　default　and　credit　risks,　and　improve　market　order.　In　this　paper,　I　propose　a　fraud　prediction　model　based　on　XGBoost.　The　data　set　required　to　build　the　model　comes　from　the　supply　chain　data　provided　by　DataGo.　Compared　with　the　model　based　on　Logistic　regression　and　the　model　of　Gausian　Naive　bayes,　the　model　proposed　in　this　paper　shows　better　classification　ability.　Specifically,　the　F1　score　based　on　the　Logistic　regression　model　is　98.96,　the　F1　score　based　on　the　Gausian　Naive　bayes　model　is　71.95,　and　the　F1　score　value　of　the　XGBoost-based　model　proposed　in　this　paper　is　99.31　in　the　experiment.　©　2021　IEEE.