Evaluating Machine Learning Techniques on Human Activity Recognition Using Accelerometer Data

Abstract

Human activity recognition is gaining increasing importance because of its implication in remote monitoring application including security, health and fitness apps. This paper provides an analysis of different machine learning techniques for recognizing human activity. Firstly, all the recent work related to human activity recognition using accelerometer data is analyzed and presented in the paper. In this study the accelerometer used in smartphones as well as those embedded in wearable devices are compared and recognition methodologies applied on both the devices are presented. The dataset used in this project is a transformed version of "Activity Recognition using Cell Phone Accelerometers," by the Wireless Sensor Data Mining WSDM. Some important features were extracted from the data and based on it different models were assessed using Matlab Classification Learner App. Four distinct machine learning techniques were applied on the dataset, namely, linear regression, logistic regression, support vector machine and neural network. For the purposed of applying classifier Weka tool is used. The results of these algorithms are compared and presented in the form of tables and graphs and Bagged Tree is identified to be the best algorithm based on accuracy results.