Recursive Parameter Estimation of Generalized Dirichlet Hidden Markov Models: Application to Occupancy Estimation in Smart Buildings

Hidden Markov model (HMM) is a classic machine learning technique to model sequences. Analyzing the characteristics of this model has been extensively studied in the past. In this paper we go through parameter estimation of HMM. We apply recursive technique in order to be able to handle real time data without suffering from extensive time complexity and memory usage in calculation. In this context, we investigate recursive parameter estimation of generalized Dirichlet (GD) HMM via the expectation-maximization (EM) framework. The GD HMM is shown to be an interesting alternative to the Dirichlet HMM. Extensive simulations based on synthetic and real data to estimate occupancy in smart buildings show the effectiveness of the recursive approach for parameter estimation.