On a two-level multiclassifier system with error correction applied to the control of bioprosthetic hand

The paper presents an advanced method of recognition of patient's intention to move of hand prosthesis. The proposed method is based on two-level multiclassifier system (MCS) with homogeneous base classifiers dedicated to EEG, EMG and MMG biosignals and with combining mechanism using a dynamic ensemble selection (DES) scheme and probabilistic competence function. Additionally, the feedback signal derived from the prosthesis sensors is applied to the correction of classification algorithm. The performance of MCS with proposed competence function and combining procedure were experimentally compared against three benchmark MCSs using real data concerning the recognition of six types of grasping movements. The systems developed achieved the highest classification accuracies demonstrating the potential of multiple classifier systems with multimodal biosignals for the control of bioprosthetic hand.