Time-resolved X-ray densitometry void fraction measurements and accompanying acoustic emissions have revealed that partial cavity shedding on a hydrofoil can be multimodal, with spontaneous changes in shedding sequence (referred to here as cavitation style) for fixed inlet flow conditions. These spontaneous, intermittent transitions between two physically different shedding styles are examined using data-based image analysis of the cavity flow fields in order to extract the associated physical mechanism leading to style transition. Three data-driven decomposition techniques are compared: proper orthogonal decomposition (POD), dynamic mode decomposition (DMD), and cluster-based reduced-order modeling (CROM), with a primary focus on the latter. The results highlight the utility of CROM over DMD and POD in the context of intermittent event analysis, both in terms of mode interpretability and transition mechanism identification. A frequency-based analysis of the CROM output revealed the existence of a shared harmonic between the different physical cavitation modes which gave further insight to the transition process. The data-based analysis ultimately illuminates the underlying flow mechanism that leads to the style transition, namely the key role of maximum cavity length buildup as it interacts with the vapor cloud collapse downstream of the hydrofoil.