Rule　reduction　and　rule　activation　are　two　important　directions　in　the　studies　of　improving　extended　belief　rule　base(EBRB)　inference　models.　However,　most　of　these　studies　are　still　suffering　challenges,　such　as　strong　subjectivity　of　parameters　determination　and/or　a　high　computational　complexity.　For　this　reason,　this　paper　proposes　an　improved　EBRB　inference　model,　which　is　called　CEAF-EBRB　model,　based　on　the　clustering　ensemble　and　activation　factor.　The　CEAF-EBRB　model　performs　multiple　data　clustering　analyses　on　historical　data　based　on　the　clustering　ensemble　firstly,　and　then　generates　extended　belief　rules　from　all　historical　data　in　the　unit　of　clusters.　Meanwhile,　the　activation　factor　is　used　to　modify　the　calculation　of　individual　matching　degrees　and　then　effectively　activate　consistent　rules　after　using　an　offline　way　to　initialize　the　activation　factor.　Finally,　the　feasibility　and　effectiveness　of　the　CEAF-EBRB　model　are　verified　through　solving　non-linear　function　fitting,　pattern　recognition,　and　medical　diagnosis.　The　proposed　model　can　provide　a　more　accurate　decision　support　for　decision-makers.　Copyright　©2023　Control　and　Decision.