Study On Machine Learning Method For Epilepsy EEG Classification

Epilepsy is a kind of nervous system disease caused by excessive discharge of neurons in the brain,which has the characteristics of paroxysm and repetition.The detection of epileptic seizures by EEG is of great significance for diagnosis and treatment.In clinical practice,EEG is generally interpreted and analyzed manually by doctors.This method is not only time-consuming but also easily affected by the subjective factors of doctors.There are some cases of delayed diagnosis,misdiagnosis,and missed diagnosis.Therefore,the research on the automatic detection method of seizures has a certain practical value.First of all,the EEG and its characteristics are described,and the feature extraction methods of EEG are analyzed from four aspects: time domain,frequency domain,timefrequency domain,and nonlinear dynamics.the advantages and disadvantages of several widely used EEG feature extraction methods are compared,and the seizure detection and classification methods of machine learning are summarized,which lays a theoretical foundation for the subsequent design of automatic seizure detection methods.Secondly,a feature extraction method based on multi-domain joint analysis is constructed.Combined with bispectrum and continuous wavelet transform algorithm,the dynamic analysis of the subband of brain telecommunication signal is carried out effectively in order to obtain the features with high distinguish.then the tree-based feature selection algorithm is used to determine the features with the greatest contribution and multi-layer perceptron is used for classification.In order to prove the effectiveness of the method,10-fold cross-validation was carried out using the epileptic EEG data set of the University of Bonn.The results show that the classification performance of multi-domain joint feature extraction algorithms is better than other algorithms.Finally,a time convolution network based on the attention mechanism is implemented for epileptic EEG classification.The multi-head self-attention mechanism is added to the time convolution network to extract the effective information of EEG signals and improve the classification effect.The amount of computation is reduced by using the convolutionbased sliding window for parallelization.When trained and tested in the epileptic EEG data set of the University of Bonn,100% accuracy was achieved in all two classification tasks and 98.67% accuracy in three classification tasks,indicating that this method has certain advantages in epilepsy classification.