Epileptic Seizure Detection Based on Multi-synchrosqueezing Transform and Multi-label Classification

Machine learning is a significant tool in the study of epileptic automatic detection system. This paper presents a novel method based on multi-synchrosqueezing transform (MSST) and multi-label classification (MLKNN). This proposed algorithm is developed on the basis of MATLAB software. Firstly, the processed EEG data is transformed by MSST. On the basis of synchronous compression transform, iterative redistribution is adopted to replace the original EEG data to complete pattern recognition. Then input the relative mean of the feature index and amplitude into the MLKNN classifier. Finally, the automatic detection of epileptic wave is realized through post-processing. The method was evaluated on the EEG data in Boston Children’s Hospital, and the results were satisfactory. The accuracy for both episodic and episodic EEG on the Boston database was 95.77%, with a mean sensitivity of 86.31% and specificity of 87.33%. Compared with the traditional automatic detection model based on machine learning, the performance of epilepsy detection is improved.