Electric　charge　is　the　primary　income　for　the　power　company.　However,　collecting　electric　charge　is　much　difficult　due　to　the　existence　of　the　risky　consumer　which　makes　the　huge　impact　on　the　normal　operation　and　development　of　the　company.　So　the　arrear　problem　of　the　risky　customers　has　become　one　of　the　focus　problems.　Based　on　the　gettable　electric　data　from　some　areas,　this　paper　proposed　an　integral　system　which　can　predict　risky　customers　according　to　the　various　scenarios.　In　the　system,　the　Random　Forest　(RF)　model　and　Extreme　Learning　Machine　(ELM)　model　are　integrated　that　can　effectively　analyze　the　obvious　features　of　the　risky　customers　and　predict　the　potential　risky　customers.　In　the　experiment　part,　it　has　shown　that　our　system　applied　to　arrear　risky　customers＇　prediction　has　higher　performance.　©　2015　IEEE.