Primary Health Care: Open Access

ISSN - 2167-1079

Abstract

Human Activities Recognition Via Smartphones Using Supervised Machine Learning Classifiers

Ahmed Younes Shdefat, Ahmed Abu Halimeh and Hee-Cheol Kim

This paper presents a way of detecting twelve daily physical human activities such as sitting, laying, standing, attaching to table, walking, jogging, running, jumping, pushups, stairs down, going up stairs, and cycling with acceleration and gyroscope sensors data resulted from using android smart mobile phones. An android application was developed to collect raw data from the sensors. The subjects preformed the twelve activities with smart phones where it is installed. Five of the samples had been selected as train data, while the rest ten samples selected as test data. In order to classify the subjects’ raw data, a program in Matlab R2016a was developed that applies twelve supervised classification algorithms models, and then compare between them in term of accuracy and speed factors. The twelve models are divided into two categories: Six of them under support vector machine (SVM); while the other six are under the k-nearest neighbor (k-NN). Finally, this study has the following results: The overall average accuracy rate with SVM cases is 89.79% in comparison with 87.81% for k-NN. The average speed rate is 47 seconds in SVM cases whereas it is 39 seconds in k-NN cases. With expansion of the number of activities up to 12 human actions, the result of the study showed that a good performance in terms of accuracy and speed was gained without losing an accuracy level achieved in the previous studies where maximum 8 activities were handled.

Top