IEEE Project Abstract

Electroencephalogram (EEG) signal recognitionbased on machine learning models is becoming more and moreattractive in epilepsy detection. For multiclass epileptic EEGsignal recognition tasks including the detection of epileptic EEGsignals from different blends of different background data andepilepsy EEG data and the classification of different types ofseizures, we may perhaps encounter two serious challenges: (1) alarge amount of EEG signal data for training are not available;(2) the models for epileptic EEG signal recognition are often socomplicated that they are not as easy to explain as a linear model.In this study, we utilize the proposed transfer learning techniqueto circumvent the first challenge and then design a novel linearmodel to circumvent the second challenge. Concretely, weoriginally combineγ -LSR with transfer learning to propose anovel knowledge and label space inductive transfer learningmodel for multiclass EEG signal recognition. By transferringboth knowledge and the proposed generalized label space fromsource domain to target domain, the proposed model achievesenhanced classification performance on target domain withoutthe use of kernel trick. In contrast to the other inductive transferlearning methods, the method uses the generalized linear modelsuch that it becomes simpler and more interpretable.Experimental results indicate the effectiveness of the proposedmethod for multiclass epileptic EEG signal recognition.