In　order　to　get　targeted　improving　strategies　for　better　teaching　quality,　it　is　necessary　to　establish　learning　data　model　for　specific　courses　covering　the　whole　learning　process,　and　accurate　and　stable　student　performance　classification　model　accordingly.　Drawing　on　existing　learning　data　analysis　of　online　teaching,　this　paper　proposes　process　oriented　learning　feature　model,　student　performance　classification　method,　and　corresponding　improving　strategies　analysis.　With　learning　data　from　＇Computer　Network　Fundamentals＇course,　in　which　combined　online　and　offline　teaching　methods　are　used,　learning　process　oriented　feature　model　are　defined　using　correlation　analysis　and　clustering　methods;　improved　Support　Vector　Machine　(SVM)　classification　method　with　grid　search　optimization　are　proposed　to　get　optimized　student　performance　classification　model,　which　gains　8%　improvement　on　the　accuracy　compared　to　classified　SVM　model.　Based　on　experiments,　the　accuracy　of　optimized　classification　model　is　90%,　with　6%　false　positive　rate　and　11%　false　negatives　rate,　which　are　reasonable　and　shows　the　accuracy　and　stability　of　our　model.　The　improving　strategies　analysis　for　different　students　classification　are　also　given,　which　can　provide　strong　support　for　teaching　quality　improvement.　©　Published　under　licence　by　IOP　Publishing　Ltd.