Evaluation of multi-class support-vector machines strategies and kernel adjustment levels in hand posture recognition by analyzing sEMG signals acquired from a wearable device.

One-vs-One (OVO) and One-vs-All (OVA) are decomposition methods for multi-class strategies used to allow binary Support-Vector Machines (SVM) to transform a given k-class problem into pairwise small problems. In this context, the present work proposes the analysis of these two decomposition methods applied to the hand posture recognition problem in which the sEMG data of eight participants were collected by means of an 8-channel armband bracelet located on the forearm. Linear, Polynomial and Radial Basis Function kernels functions and its adjustments level were implemented combined to the strategies OVO and OVA to compare the performance of the SVM when mapping posture data into the classification spaces spanned by the studied kernels. Acquired sEMG signals were segmented considering 0.16 s e 0.32 s time windows. Root Mean Square (RMS) feature was extracted from each time window of each posture and used for SVM training. The present work focused in investigating the relationship between the multi-class strategies combined to kernels adjustments levels and SVM classification performance. Promising results were observed using OVA strategy which presents a reduced number of binary SVM implementation achieved a mean accuracy of 97.63%.