Recognizing Human Activities in Smartphones: A Comprehensive Algorithm Assessment of a Multiple Context Approach
Lists
Only you see this
Abstract
This paper aims to assess the performance of various machine-learning algorithms for recognizing human activities in smartphones and find the best algorithm in terms of classification accuracy and computation efficiencies in both model learning and real-time activity recognizing. The underlying human activity recognition (HAR) framework works directly in mobile devices by computing the spatial contexts, temporal contexts, spatiotemporal contexts and user contexts in a real-time manner. Various supervised machine learning classification algorithms are applied under this framework for recognizing daily activities such as working, having a meeting and having a meal. The performance assessment is conducted among seven popular supervised learning schemes including Rules, Bayes, decision tree (DT), instance-based classifier (IBC), logistic regression (LR), support vector machine (SVM), and artificial neural networks (ANN). A total number of eighteen algorithms have been assessed in this study to identify their performance, focusing on both classification accuracy and computational efficiency. A live dataset has been collected for five days with three participants using smartphones and a 10-fold cross-validation strategy is applied to train the classification models of the algorithms. The test results demonstrate that the REPTree algorithm from the scheme of DT outperforms others in terms of classification computation efficiency, classification accuracy and model-learning computation efficiency.
BibTeX
@article{liu2016recognizing,
author = {Keqiang Liu and Ruizhi Chen and Tianxing Chu and Yunjia Wang and Jingbin Liu and Yuwei Chen and Ling Pei},
title = {{Recognizing Human Activities in Smartphones: A Comprehensive Algorithm Assessment of a Multiple Context Approach}},
journal = {Journal of Residuals Science and Technology},
year = {2016},
}Loading related content...