Dynamic feature selection applied to the recognition of grasping movements in the control of bioprosthetic hand

The paper presents novel method of dynamic feature selection (DFS) and its application in the problem of recognition of patient intent in the bioprosthesis control system. In the proposed approach features are selected dynamically, i.e. separately for each classified object according to the local value of usability measure of primary features. The usability measure is determined in the supervised learning procedure using randomized reference model. The performance of the DFS method was experimentally compared with four other feature selection algorithms. The approach developed achieved the highest classification accuracy demonstrating the potential of the DSF method for the control of bioprosthetic hand.