Liquid water content (LWC) governs cloud radiative forcing and precipitation efficiency, and its contrast between warm and supercooled liquid-phase clouds shapes cloud lifetime and the water cycle. Aircraft campaigns over the North China Plain have characterized droplet spectra, vertical structure, and aerosol responses, but LWC has mostly been diagnosed with single-factor or linear analyses, which cannot distinguish the individual contributions of different factors. It thus remains unclear which factors govern LWC in each regime, and whether any acts differently between them. Using in-situ observations from 13 K350 aircraft flights during 2019–2021, we compared the two cloud types and quantified predictor contributions and interactions in an LWC prediction model with SHAP. An equal-weighted AdaBoost–LightGBM ensemble performed best. Adding further members provided no improvement, indicating that model complementarity outweighs ensemble size. Despite similar cloud-base distributions, cold clouds developed more deeply with stronger winds but lower LWC. The principal vertical contrast occurred at 1600–2300 m, where warm-cloud LWC temporarily exceeded cold-cloud LWC. Wind-related dynamical factors dominated LWC predictions in both cloud types (~41 % combined). From warm to cold clouds, the temperature contribution decreased from 15.5 % to 6.9 %, while aerosol-related contributions increased. The SHAP interaction between temperature and vertical wind changed from positive to negative, indicating that the two regimes share a dynamic framework in which the temperature-wind coupling, not temperature itself, differs. By making these contributions directly comparable, the study identifies this interaction reversal as the key regime difference and provides an observational constraint for LWC parameterizations.