Discretization　is　an　important　algorithm　and　considered　to　be　a　process　of　information　generalization　and　data　reduction.　To　avoid　information　loss　and　total　number　of　cut　point　decrease　after　discretization　of　continuous　attributes,　based　on　multi-attribute　discretization　algorithm　with　good　global　clustering　effects　for　selecting　candidate　cut　points　is　proposed.　The　improved　algorithm　is　combined　with　the　advantages　of　clustering　method　and　algorithm　based　on　the　importance　of　cut　points,　The　experimental　results　show　that　the　proposed　algorithm　can　significantly　decrease　the　number　of　discretization　cut　points　and　increase　the　predictive　accuracy　of　the　classifier　than　both.