An Improved Hybrid Algorithm for Optimizing the Parameters of Hidden Markov Models

Abstract

Hidden Markov Models (HMMs) have become increasingly popular in the last several years due to the fact that, the models are very rich in mathematical structure and hence can form the theoretical basis for use in a wide range of applications. Various algorithms have been proposed in literature for optimizing the parameters of these models to make them applicable in real-life. However, the performance of these algorithms has remained computationally challenging largely due to slow/premature convergence and their sensitivity to preliminary estimates. In this paper, a hybrid algorithm comprising the Particle Swarm Optimization (PSO), Baum-Welch (BW), and Genetic Algorithms (GA) is proposed and implemented for optimizing the parameters of HMMs. The algorithm not only overcomes the shortcomings of the slow convergence speed of the PSO but also helps the BW escape from local optimal solution whilst improving the performance of GA despite the increase in the search space. Detailed experimental results demonstrates the effectiveness of our proposed approach when compared to other techniques available in literature.