The　extended　belief　rule　base　(EBRB)　system　has　been　successfully　applied　to　classification　problems　in　various　fields.　However,　the　existing　EBRB　generation　method　converts　all　data　into　extended　belief　rules,　which　leads　to　the　large　scale　of　rule　base　and　affects　the　efficiency　and　accuracy　of　subsequent　inference.　In　view　of　this,　this　paper　proposes　an　EBRB　rule　reduction　method　based　on　the　adaptive　K-means　clustering　algorithm　(RC-EBRB).　In　the　rule　generation　process,　the　K-means　clustering　algorithm　is　applied　to　obtain　the　rule　cluster　centers,　which　are　used　to　generate　new　rules.　In　the　end,　these　new　rules　form　a　reduced　EBRB.　Moreover,　in　order　to　determine　the　initial　cluster　centers　and　the　number　of　clusters　in　the　K-means　clustering　algorithm,　the　algorithm　idea　of　K-means++　is　introduced　and　a　reduction　granularity　adjustment　algorithm　with　threshold　is　proposed,　respectively.　Finally,　four　datasets　on　commonly　used　classification　datasets　from　UCI　are　used　to　verify　the　performance　of　the　proposed　method.　The　experimental　results　are　compared　with　the　existing　EBRB　methods　and　the　traditional　machine　learning　methods,　which　prove　the　effectiveness　of　the　method.　©　2021,　Springer　Nature　Switzerland　AG.