The spatiotemporal characteristics of fluidization within an Air Dense Medium Fluidized Bed (ADMFB) are critical for its heat and mass transfer, as well as for mixing and separation processes. Yet, detailed analyses of the dynamic behavior of flow regimes during progressive transformations are scarce. This study captured synchronous pressure drop signals at multiple points along the sidewall of the bed using a non-intrusive measurement technique. Statistical analyses were utilized to extract the pressure drop's characteristic parameters and examine the statistical features' spatiotemporal fluctuations during fluidization state transitions. We proposed a method to quantitatively characterize the bed's real-time fluidization state through the combined use of multiple-point pressure characteristics. Following this approach, a multidimensional spatio-temporal matrix was constructed to depict the flow regime transitions quantitatively. Concurrently, ensemble learning techniques were applied to classify flow regimes based on various feature parameters and sample lengths. In the final phase, comprehensive interpretations and detailed visual analyses were deployed on the inputs to the optimized ensemble learning model, XGBoost, utilizing the Shapley Additive Explanations (SHAP) framework. This study provides additional insights and methodologies for managing the transition and intelligent control of flow regimes within an ADMFB.