With　the　increasing　demand　of　bank　loan　businesses,　the　probability　of　non-performing　loans,　that　is,　loan　default,　has　also　increased　sharply.　We　design　machine　learning　algorithm　to　solve　the　problem,　which　can　reduce　the　loan　risk　and　improve　service　efficiency,　especially　when　we　face　the　data　unbalanced　issues.　Firstly,　we　train　the　random　forest　model　with　the　historical　bank　loan　data　and　associated　data　from　other　financial　institutions.　Secondly,　we　revised　the　unbalanced　data　classification　algorithm　with　random　forest　and　tuned　the　data　feature　extraction　methods.　Thirdly,　the　results　show　that　the　machine　learning　risk　predication　algorithm　outperforms　traditional　statistical　algorithms.　In　addition,　we　use　random　forest　algorithm　to　identify　the　impact　of　data　feature,　it　is　possible　to　obtain　features　that　have　a　huge　impact　on　the　definition　of　the　results,　allowing　for　more　accurate　loan　risk　assessment　in　the　financial　sector.　©　2022　ACM.