A Novel Double-Neighborhood Method for Robust Epileptic Seizure Classification

This is the authors' abstract. We don't add key points for this paper.

The realm of Machine Learning techniques is ever-evolving, with new algorithms and innovative combinations of existing approaches being utilized to achieve outstanding classification performance in various fields. In this series, the proposed method used the concepts of double neighbourhood, weight, and distance. Firstly, a -neighborhood is established using k points in the n-dimensional space, within which K clusters are envisioned, where K is equal to the number of unique target classes. To reduce the effect of noise and eliminate skewness, another neighborhood is generated with Euclidean distance and centroid for each cluster i. Then, weight of each cluster is evaluated to identify the class of the unknown test instance (z). To evaluate the effectiveness of this method, experimental analyses are conducted using two widely utilized datasets: the CHB-MIT dataset and the BONN dataset. The proposed method achieved 98.0% accuracy for the CHB-MIT dataset and 97.89% in predicting two-class classification, 96.21% in detecting three-class classification and 94.61% in distinguishing five-class classification on the BONN dataset. Compared with other renowned methods, namely, traditional KNN, Support Vector Machine(SVM), Naïve Bayes(NB), Logistic Regression(LR), and Decision Tree(DT), the proposed model recorded good performance in all the levels of classification for both the datasets and is advantageous by effectively handling binary as well as multiclass classification.