Precise　and　timely　detection　of　anomalous　modeling　behaviors　is　critical　for　both　teaching　and　application　of　modeling　methods.　Existing　methods　usually　focus　on　evaluating　the　modeling　results　rather　than　mining　the　knowledge　hidden　in　the　modeling　process.　In　this　paper,　we　propose　to　monitor　and　analyze　the　modeling　process　in　order　to　timely　detect　anomalous　modeling　behaviors,　potentially　contributing　to　a　comprehensive　assessment　of　the　modeling　practice.　Specifically,　we　propose　to　systematically　build　a　goal　model　for　characterizing　the　normal　modeling　behaviors,　which　establishes　the　connections　between　modelers＇　high-level　modeling　behaviors　and　low-level　modeling　operations.　On　top　of　such　a　goal　model,　we　propose　a　data　mining-based　approach　to　semi-Automatically　validate　the　design　of　the　goal　model　and　explore　other　normal　modeling　behaviors.　Then,　we　propose　to　automatically　detect　anomalous　modeling　behaviors　by　capturing　normal　modeling　behaviors　obtained　from　the　goal　model　and　actual　modeling　sequences.　We　have　developed　and　deployed　a　data-flow　diagram　modeling　platform,　which　implemented　our　proposed　approach.　We　have　conducted　an　experiment　with　57　participants,　the　preliminary　results　of　which　show　that　our　approach　can　effectively　detect　modelers＇　anomalous　behaviors.　The　experiment　results　are　beneficial　for　not　only　assessing　the　modelers＇　performance　but　also　identifying　the　usability　issues　of　the　modeling　tool.　©　2022　ACM.