Human activity recognition using supervised machine learning techniques

Human activity recognition (HAR) is one of the common classification problems that has a lot of interest. In HAR, different techniques are explored to identity the physical activity of humans, given data coming from various sensors. This paper explored the use of different supervised learning techniques and algorithms such as Gaussian Bayes Theory, Logistics Regression, SVM Classifier, Quadratic Discriminant Analysis, k-Nearest Neighbor and random Forest Classifier. Accuracy measurement of how each algorithm successfully predict the correct classification is done with the Random Forest Classifier giving the highest user accuracy of 87% without cross validation and 84.9% with 5-fold cross validation. For class accuracy, Random Forest Classifier gives an accuracy of 90.5% without cross validation while k-nearest neighbor gives an accuracy of 89.2% with k=3 and 5-fold cross validation.