Computational Framework for Temporal State Recognition: HMM-Based Modeling and Prediction of Discrete Behavioral Sequences

This study addresses the computational challenge of mapping temporal behavioral sequences to latent states in multi-class classification scenarios. We propose a Hidden Markov Model (HMM)-based framework that models observation–state transitions for sequential event streams, enabling automated inference of unobservable states from discrete behavioral features. The approach incorporates state transition modeling, observation probability estimation, and optimal sequence decoding via the Viterbi algorithm. Experiments on a domain-specific dataset demonstrate an overall classification accuracy of 82.3%, with the highest recognition performance (91.5%) in the most distinctive state category. Compared with conventional threshold-based methods, the proposed framework achieves higher cross-class generalization and improved robustness to feature overlap. This work provides a standardized, model-driven solution for temporal state recognition tasks, with potential applicability to other cross-domain behavioral sequence analysis problems.