Title: Human activity recognition based on mobile phone sensor data using stacking machine learning classifiers
Authors: Mahsa Soufineyestani; Hedieh Sajedi; Vali Tawosi
Addresses: Computer Science Department, University of Minnesota, Duluth, USA ' School of Mathematics, Statistics and Computer Science, College of Science, University of Tehran, Tehran, Iran ' Electrical and Computer Engineering Department, Tarbiat Modares University, Tehran, Iran
Abstract: Human activity recognition aims to determine which activity is performed by individuals. It has plenty of real-world applications such as health monitoring and abnormal behaviour detection. Therefore, this study focuses on distinguishing and classifying human activities by applying statistical features and using stacking learning methods with the aim of improving the accuracy and precision of the classification. At first, features are extracted from raw sensor data and 26 subsets of the complete feature set are determined and tested to see which subset results in a higher precision. Then a feature selection technique based on the genetic algorithm is applied to the extracted features to observe if it can improve the results. Comparative results between classifiers showed that stacking models have advantages in increasing classification accuracy, especially in the case of climbing stairs and walking that are difficult to distinguish by single classifiers.
Keywords: activity recognition; classification; feature extraction; stacking learning algorithm; feature selection.
International Journal of Digital Signals and Smart Systems, 2019 Vol.3 No.4, pp.204 - 232
Received: 07 Oct 2018
Accepted: 06 Dec 2018
Published online: 04 Mar 2020 *