Antarctic sea ice has undergone unprecedented changes in recent years, marked by a stepwise shift from long-term expansion to rapid decline since 2016. Yet the underlying processes remain poorly understood due to the complex interplay of dynamic and thermodynamic drivers. Using an ascendant data mining framework, we objectively classify sea ice into distinct process-based regimes during the growth season in the NEMO-SI<sup>3</sup> model. Six robust regimes with unique combinations of dynamic and thermodynamic drivers are identified. We reveal that a coastal regime, characterized by strong dynamic divergence and compensating thermodynamic ice growth, has experienced significant area loss since 1979, accelerating after 2016, potentially reflecting weakened new ice formation. Meanwhile, sea ice extent anomalies closely correlate with both the coastal and a pack ice regime, highlighting their dominant role in overall sea ice variability. By capturing physically consistent sea ice regimes, our work offers a new process-based perspective for understanding Antarctic sea ice variability and its unprecedented recent changes.