To　automatically　construct　a　Mamdani　fuzzy　model,　a　novel　approach　is　proposed　based　on　local　density　of　data.　The　fuzzy　rule　base　and　membership　function　parameters　for　a　candidate　fuzzy　system　can　be　determined　through　the　data　mining　using　the　clustering　algorithm　of　fuzzy　clustering　of　local　approximation　of　membership　(FLAME),　and　consequently　the　fuzzy　system　is　generated　automatically.　In　the　clustering　process,　there　is　no　requirement　to　specify　the　number　of　clusters　and　the　outliers　can　be　automatically　identified　without　any　extra　pre-processing.　The　proposed　approach　is　evaluated　through　a　set　of　simulated　experiments　on　the　traffic　prediction　and　the　results　indicate　that　the　proposed　approach　for　fuzzy　system　identification　is　feasible　and　efficient.