Abstract
This paper proposed a novel algorithm which named Randomized Particle Swarm Optimization (RPSO) to optimize HMM for human activity prediction. The experiments designed in this paper are the classification of human activity using two data sets. The first testing data is from the TUM Kitchen Data Set and the other is the Human Activity Recognition using the Smartphone Data Set from UCI Machine Learning Repository. Based on the comparison of the accuracies for the conventional HMM and optimized HMM, a conclusion can be drawn that the proposed RPSO can help HMM to achieve higher accuracy for human action recognition. Our results show that RPSO-HMM can improve 15% accuracy in human activity recognition and prediction when compared to the traditional HMM.