An Overview of Classification Techniques for Human Activity Recognition

In this paper, both classic and less commonly used classification techniques are evaluated in terms of recognizing human activities recorded in the PAMAP2 dataset that was created using three inertial measurement units. Seven algorithms are compared in terms of their accuracy performance with the best classifier being based on the Orthogonal Matching Pursuit algorithm that has been modified to remove the limitation of the number of training vectors per class present in its original version. The overview shows that human activities as defined by the PAMAP2 dataset can be recognized reliably even without any prior data preprocessing.