An Extended K Nearest Neighbors-Based Classifier for Epilepsy Diagnosis

In the diagnosis of epileptic seizures, classification is an important step that directly affects the results. Visual inspection of Electroencephalogram (EEG) is a relatively common analytic method of epilepsy, but it is costly, time-consuming and relies on the experiences of the doctor. Therefore, the development of an efficient and accurate epileptic seizure automatic diagnosis system suitable for clinical diagnosis has become an urgent task. In order to better solve the problem of early diagnosis of epileptic and bring timely treatment to patients, the comprehensive representation of k nearest neighbors for multi-distance decision making (CRMKNN) is proposed in this research. In the proposed scheme, Euclidean distance and Hassanat distance are firstly used to select neighbors. Subsequently, the similarity distance is obtained through the linear representation of the nearest neighbors, and calculate the distribution of nearest neighbors in the category to get the discrete distance. Finally, the distance based on the comprehensive representation of the category is used to determine the category of the query EEG signal. In order to verify the method, we used the EEG signals from Bonn university public database and conducted experiments on six kinds of EEG combinations. Experimental results showed that our method could automatically detect seizure in all situations with an accuracy of not less than 99.50%. At the same time, compared with the classification results of existing methods, this method is more effective.