Personal　loans　are　the　core　and　foundation　of　financial　business.　Traditional　personal　credit　approval　is　performed　manually.　With　the　large-scale　popularization　of　the　Internet,　users　can　conveniently　apply　for　loans　through　various　financial　platforms.　Personal　credit　has　the　characteristics　of　large　crowd　size,　small　amount,　short　cycle,　and　high　complexity.　Manual　review　is　not　only　inefficient　but　also　has　low　accuracy　in　predicting　user　default.　Credit　default　technology　based　on　machine　learning　algorithms　emerged　as　the　times　require.　However,　the　credit　default　dataset　is　a　typical　imbalanced　dataset.　The　imbalanced　dataset　seriously　affects　the　performance　of　the　algorithm,　which　causes　the　algorithm　to　only　focus　on　majority　class　samples　and　ignore　minority　class　samples.　This　paper　proposes　an　algorithm　called　multi-granularity　class　weighting　that　can　effectively　handle　the　class　imbalance　problem.　The　main　contribution　of　this　paper　is　to　propose　an　algorithm　called　multi-granularity　class　weighting　to　deal　with　the　class　imbalance　problem,　propose　a　new　feature　generation　algorithm　to　generate　high-quality　features,　and　use　the　stacking　method　in　ensemble　learning　to　improve　the　model　performance.　©　2024　IEEE.