Objective The k-nearest neighbor algorithm is used to calculate the partial mutual information of symbol sequence to analyze the coupling relationship between epileptic EEG signals, to explore the difference in the coupling degree between epileptic EEG signals and healthy people's EEG signals, and to provide a reference method for the study of epileptic EEG signals. Methods The traditional method is to obtain partial mutual information by calculating the probability distribution density between variables. In this paper, the k-nearest neighbor method is used to calculate part of mutual information. The algorithm does not require much data, and the accuracy and efficiency of the algorithm are relatively high. First, the original epileptic EEG signal sequence is symbolized. The purpose of symbolization is to transform the sequence into a symbol sequence, which can effectively reduce the impact of noise. Then, the symbol sequence is coded. Finally, the k-nearest neighbor algorithm is used to calculate partial mutual information to obtain the coupling relationship of EEG signals. Results For traditional methods of obtaining partial mutual information, when the data length is greater than 4000, the p-values obtained in the pillow regions O1 and O2 are less than 0.05. For the k-nearest neighbor method to obtain partial mutual information, when the data length is greater than 2000, the p-values obtained in the pillow regions O1 and O2 are less than 0.05. Compared to traditional methods, the k-nearest neighbor method can distinguish between epileptic EEG signals and healthy EEG signals in experimental data using a shorter data length. At the same time, it was found that the coupling degree of EEG signals in healthy individuals was significantly higher than that in epilepsy patients.. Conclusions The k-nearest neighbor method can effectively analyze epileptic EEG signals. The accuracy and efficiency of the algorithm are relatively high.