Continuous Behavioral State Modeling via HMM for Adherence Recognition in Psychoeducational Intervention Processes

Current compliance identification methods in psychoeducational interventions often rely on discrete retrospective assessment and cannot dynamically capture continuous state evolution. To address this limitation, this paper proposes a continuous behavioral state modeling method based on the Hidden Markov Model. Real-time user interaction behavior sequences are collected through a digital intervention platform and used as observation data. Potential internal compliance states are defined according to psychoeducational theory, and the probabilistic relationship between observed sequences and latent states is modeled using HMM. Through parameter learning of state transition and observation probabilities and decoding algorithms, the user’s real-time compliance state sequence is dynamically inferred, enabling continuous and automated tracking of compliance levels. Experimental results show that the proposed method is highly consistent with expert annotations in identifying turning points, achieving a state sequence alignment accuracy of 0.833, outperforming random forest and rule-based methods. For early warning, the model detects “significantly decreased compliance” and “intervention dropout” events 12.5 and 18.2 days earlier on average, with recall of 0.900 and AUC of 0.892.