Research On Classification Method Of Epilepsy EEG Data Based On Machine Learning

Epilepsy is a relatively common brain disease in the world.This neurological disease can directly threaten the life safety of patients during seizures.Therefore,rapid clinical diagnosis of epilepsy is an effective way to help patients reduce the risk.EEG can reflect important information of brain discharge and is an important tool for evaluating brain dysfunction such as epilepsy.Visual EEG is a commonly used epilepsy analysis method,but this method is undoubtedly very dependent on the experience of doctors,and it is very time-consuming,and there is the possibility of misdiagnosis and missed diagnosis.In view of this,it is an urgent task to develop an efficient,accurate and clinically applicable epileptic seizure diagnosis system.In this thesis,based on machine learning,two improved K Nearest Neighbor(KNN)algorithms are proposed to achieve various classification tasks on epileptic EEG signals,and the proposed method is analyzed using the EEG database of the University of Bonn,Germany.The main research contents of this thesis are as follows:(1)For the multi-classification problem of epilepsy,a weighted K-nearest neighbor classifier based on Bray Curtis distance(WBCKNN)is proposed.The method uses discrete Fourier transform to convert the time domain features of the signal into frequency domain features,and uses the Bray Curtis distance to measure the sample similarity,so as to obtain more disease information contained in the signal.This improves the k-value sensitivity problem of the KNN algorithm.At the same time,concepts such as local mean vector and generalized mean distance are introduced.By adjusting the parameters,the contribution of samples with high similarity to classification can be amplified to the greatest extent.In the experiment,the effectiveness is compared with six other machine learning methods.The results showed that the proposed method has the best classification accuracy of 99.67% and 99% for the two-class and three-class problems,and the sensitivity and specificity are better than the traditional methods.(2)Aiming at various classification tasks during epileptic seizures,a multi-distance decision classifier based on K-nearest neighbor comprehensive representation(CRMKNN)is proposed.This method uses the combination of Euclidean distance and Hassan distance for neighborhood selection,and then obtains the similarity distance through the linear representation of the nearest neighbor.And calculate the distribution of the nearest neighbors in the category to obtain the discrete distance,which effectively improves the problem that the KNN method is susceptible to outliers and parameters.At the same time,in order to effectively verify the difference between the CRMKNN method and other KNN algorithms,seven other commonly used KNN improved algorithms were compared.The results show that the combined accuracy of the two-classification of the six epilepsy signals is higher than 99.50%,and the best classification accuracy for the three-classification problem is 98.33%.And as the value of k increases,the classification error rate gradually decreases and eventually tends to be stable.