Activity recognition using smartphone embedded sensors with k-nearest neighbor algorithm

Human Activity Recognition(HAR) is behind many human-centric applications such as Smart Homes, and public healthcare. This paper presents a way of detecting six basic physical human activities (Sitting, Laying, Standing, Walking, Walking Downstairs, Walking Upstairs), with accelerometer embedded in an android smartphone. K-Nearest Neighbor (KNN) algorithm, and Relief F feature selector in WEKA Machine Learning Toolkit were adopted to recognize these activities on a publicly available dataset. The average recognition accuracy of 95.2% demonstrated the feasibility of the proposed solution, and can be considered adequate for HAR. This work is significant because it can provide contextual information about the habits of users passively.