Evaluating state-of-the-art classifiers for human activity recognition using smartphones

Human activity recognition using smartphones and wearables is a field gathering a lot of attention. Although a plethora of systems have been proposed in the literature, comparing their results is not an easy task. As a universal evaluation framework is absent, direct comparison is not feasible. This paper compares state-of-the-art classifiers already used on mobile human activity recognition, under the same conditions. In addition, an Android application was developed and the method yielding the best results was evaluated in real world in a semi-supervised environment. Results shown that deep learning techniques have better performance and could be transferred to a phone without many modifications.