Application of a novel feature selector for human activity recognition based on inertial monitored data

摘要

The last technological advances in wearable sensors and machine learning are allowing for a new generation of human monitoring techniques, with an especial interest for the analysis of human biomechanics and activity recognition. In this paper, an application of intelligent systems to solve the problem of daily physical activity recognition is presented. Taking into account the importance of data featuring and the selection of the most important features for subsequent pattern recognition stage, a new feature selection methodology based on a filter technique via a couple of two statistical criteria is presented. Satisfactory accuracy rates are achieved by using support vector machines especially for preprocessed data, with remarkable accuracy and applicability in the case of the wrist location.