A fast convergent and robust classifier for multi-way corrupted eeg signals

The performance of the statistical classifiers being employed determines how precisely an electronic prosthesis moves its target. The performance of the classifier is correlated with the quality of the acquired data, and it degrades in the presence of incomplete data. Missing data must therefore be recovered in order to enhance classifier performance and subsequently regulate prosthesis performance. However, recovering lost data can be difficult. For the purpose of recovering missing data in electroencephalography (EEG) signals, a fast and accurate Xavier-initialized weighted canonical polyadic (X-WCP) based approach has been described in this study. In comparison to cutting-edge techniques, the Xavier-initialized weights converge quickly since they are close to the optimal solution. This research also shows the effectiveness of the proposed approach for enhancing classifier performance. For the method to be validated, three models are proposed. Classifiers are employed to complete EEG data in the first model, and the results are used as a benchmark. In the second and third models, classifiers are employed on partial and recovered signals. The results demonstrate that when recovered data is provided to classifiers instead of partially complete EEG data, where classification accuracy was 55%, classification accuracy is greatly enhanced (up to 70%). Classification accuracy at the benchmark level was 71%.