FEATURE SELECTIONS FOR HUMAN ACTIVITY RECOGNITION IN SMART HOME ENVIRONMENTS

In this paper, three probabilistic models are applied to represent and recognize human activities from observed sensor sequences: Naive Bayes classifier, forward procedure of a Hidden Markov Model and Viterbi algorithm based on a Hidden Markov Model. A variety of different feature selection methods is tested in order to reduce the dimensionality of the learning problem. The results show that the activity recognition performance measures of the three algorithms have a strong relationship with the dataset features that are utilized. Larger time feature values and smaller length size feature values will generate better results, relatively.