Applying Semi-Supervised learning on Human Activity Recognition Data

Abstract

Mobile devices have now become a part of our lives, and due to the sophisticated sensors incorporated in them, their utility to the users is enormous. At the same time the data thus captured can be analyzed for applications ranging from user identification, user authentication to activity monitoring. Machine learning algorithms for supervised classification are restricted by their requirement of large amount of labelled data. In such cases supervised learning may be adopted as it uses labelled as well -semi as unlabeled data to perform the learning tasks and construct better classifiers. In this paper semi supervised learning approach has been applied on the accelerometer data (labelled as well as unlabeled), collected from smart phones, for human activity recognition. The results have been compared with supervised classifiers and are found to be comparable.