A Study on the Effect of Features Obtained From Signal Segments on Classification Success

Successful classification depends on the selection of distinctive features and effective channel subset used in classification. In this study, novel practical methods are proposed for determining detecting subsets multi systems such as EEG. Two different feature extraction compared study. The first one is based classical Wavelet transform second our approach which slope signal segments. Feature vectors generated from some properties mean, standard deviation, numerical integral coefficients method. For method, only slopes segments vectors. Signal Path Slope (SPS) differently a Savitzky Golay (S-G) filter with an optimal frame length applied to before segmentation make path more prominent time domain. way, extracted by using S-G filter. selection, iterative method results divide dataset labelled into two groups % 90 pre-training %10 pre-test data proposed. provided dataset-3 BCI competition IV classified each Support Vector Machine classifier. given comparatively it observed that has less computational complexity successful than methods. highest accuracies 67.74 49.27 subject-1 subject-2 respectively obtained low dimensional vector SPS achieved study increased 8.24 14.97 when average results. significant increase success both subjects shows consistency By there subject-specific pattern related motor imagery tasks brain. This successfully determined