Onder Aydemir1, Ebru Ergün2. 1. Department of Electrical and Electronics Engineering, Karadeniz Technical University, 61080, Trabzon, Turkey. Electronic address: onderaydemir@ktu.edu.tr. 2. Department of Electrical and Electronics Engineering, Recep Tayyip Erdoğan University, 53100, Rize, Turkey. Electronic address: ebru.yavuz@erdogan.edu.tr.
Abstract
BACKGROUND: The input signals of electroencephalography (EEG) based brain computer interfaces (BCI) are extensively acquired from scalp with a multi-channel system. However, multi-channel signals might contain redundant information and increase computational complexity. Furthermore, using only effective channels, rather than all channels, may enhance the performance of the BCI in terms of classification accuracy (CA). NEW METHOD: We proposed a robust and subject-specific sequential forward search method (RSS-SFSM) for effective channel selection (ECS). The ECS procedure executes a sequential search among each of the candidate channels in order to find the channels which maximize the CA performance of the validation set. It should be noted that in order to avoid the problems of random selections in the validation set, we applied the ECS procedure for 100 times. Then, the total numbers of the selection of each channel present the effective ones. To demonstrate its reliability and robustness, the proposed method was applied to two data sets. RESULTS: The achieved results showed that the proposed method not only improved the average CA by 15.98%, but also decreased the considered number of channels and computational complexity by 71.53% on average. COMPARISON WITH EXISTING METHOD(S): Compared with the existing methods, we achieved better results in terms of both the classification accuracy improvement and channel reduction rates. CONCLUSIONS: Features extracted by Hilbert transform and sum derivative methods were effectively classified by support vector machine. In conclusion, the results obtained proved that the RSS-SFSM shows great potential for determining effective channel(s).
BACKGROUND: The input signals of electroencephalography (EEG) based brain computer interfaces (BCI) are extensively acquired from scalp with a multi-channel system. However, multi-channel signals might contain redundant information and increase computational complexity. Furthermore, using only effective channels, rather than all channels, may enhance the performance of the BCI in terms of classification accuracy (CA). NEW METHOD: We proposed a robust and subject-specific sequential forward search method (RSS-SFSM) for effective channel selection (ECS). The ECS procedure executes a sequential search among each of the candidate channels in order to find the channels which maximize the CA performance of the validation set. It should be noted that in order to avoid the problems of random selections in the validation set, we applied the ECS procedure for 100 times. Then, the total numbers of the selection of each channel present the effective ones. To demonstrate its reliability and robustness, the proposed method was applied to two data sets. RESULTS: The achieved results showed that the proposed method not only improved the average CA by 15.98%, but also decreased the considered number of channels and computational complexity by 71.53% on average. COMPARISON WITH EXISTING METHOD(S): Compared with the existing methods, we achieved better results in terms of both the classification accuracy improvement and channel reduction rates. CONCLUSIONS: Features extracted by Hilbert transform and sum derivative methods were effectively classified by support vector machine. In conclusion, the results obtained proved that the RSS-SFSM shows great potential for determining effective channel(s).