Considering　the　characteristics　of　longitudinal　data　set,　such　as　multi-variates,　missing　data,　unequal　series　length,　and　irregular　time　interval,　an　algorithm　based　on　Eros　distance　similarity　measure　for　longitudinal　data　is　proposed.　Eros　distance　is　used　in　Fuzzy-C-Means　cluster　processing.　First,　preprocessing　is　done　for　unbalance　longitudinal　data　set,　which　includes　filling　the　missing　data,　reducing　the　randaut　attributes,　etc.　Second,　FErosCM　Cluster　method　is　used　for　classification　automatically,　and　takes　into　account　information　entropy　for　assessing　the　performance　of　cluster　algorithm.　Experiments　show　that　this　method　is　effective　and　efficient　for　longitudinal　data　classification.