Physical activity classification using resilient backpropagation (RPROP) with multiple outputs

Abstract

Considerable research has been conducted into the classification of Physical activity monitoring, an important field in computing research. Using artificial neural networks model, this paper explains novel architecture of neural network that can classify physical activity monitoring, recorded from 9 subjects. This work also presents a continuation of benchmarking on various defined tasks, with a high number of activities and personalization, trying to provide better solutions when it comes to face common classification problems. A brief review of the algorithm employed to train the neural network is presented in the first section. We also present and discuss some preliminary results which illustrate the performance and the usefulness of the proposed approach. The last sections are dedicated to present results of many architectures networks. In particular, the experimental section shows that multiple-output approaches represent a competitive choice for classification tasks both for biological purposes, industrial etc.