Among　many　rule-based　systems　employed　to　deal　with　classification　problems,　the　extended　belief-rule-based　(EBRB)　system　is　an　effective　and　efficient　tool　and　also　has　potentials　in　handing　both　quantitative　and　qualitative　information　under　uncertainty.　Despite　many　advantages,　several　drawbacks　must　be　overcome　for　better　applying　the　conventional　EBRB　system,　including　counterintuitive　individual　matching　degrees,　insensitivity　to　the　calculation　of　individual　matching　degrees,　and　the　inconsistency　problem.　Accordingly,　by　constructing　the　activation　region　of　extended　belief　rules　and　revising　the　calculation　formula　of　activation　weights,　the　new　procedures　of　activation　rule　determination　and　weight　calculation　are　proposed　to　improve　the　conventional　EBRB　system,　while　the　original　procedures　of　rule　inference　and　class　estimation　are　retained　from　the　conventional　EBRB　system.　Nineteen　classification　datasets　with　different　numbers　of　classes　are　studied　to　validate　the　efficiency　and　effectiveness　of　the　proposed　EBRB　classification　system　compared　with　existing　works.　The　comparison　results　demonstrate　that　the　proposed　EBRB　classification　system　not　only　obtains　a　high　accuracy　better　than　the　conventional　EBRB　system,　but　also　has　an　excellent　response　time　for　classification.　More　importantly,　the　results　derived　from　multi-class　datasets　show　the　significant　performance　of　the　proposed　EBRB　classification　system　compared　with　some　state　of　art　classification　tools.　©　2018　Elsevier　B.V.