Epileptic Seizure Detection in EEGs by Using Random Tree Forest, Naïve Bayes and KNN Classification

Abstract

Epilepsy is a disease that attacks the nerves. To detect epilepsy, it is necessary to analyze the results of an EEG test. In this study, we compared the naive bayes, random tree forest and K-nearest neighbor (KNN) classification algorithms to detect epilepsy. The raw EEG data were pre-processed before doing feature extraction. Then, we have done the training in three algorithms: KNN Classification, naïve bayes classification and random tree forest. The last step was validation of the trained machine learning. Comparing those three classifiers, we calculated accuracy, sensitivity, specificity, and precision. The best trained classifier is KNN classifier (accuracy: 92.7%), rather than random tree forest (accuracy: 86.6%) and naïve bayes classifier (accuracy: 55.6%). Seen from precision performance, KNN Classification also gives the best precision (82.5%) rather than Naïve Bayes classification (25.3%) and random tree forest (68.2%). But, for the sensitivity, Naïve Bayes classification is the best with 80.3% sensitivity, compare to KNN 73.2% and random tree forest (42.2%). For specificity, KNN classification gives 96.7% specificity, then random tree forest 95.9% and Naïve bayes 50.4%. The training time of naïve bayes was 0.166030 sec, while training time of random tree forest was 2.4094sec and KNN was the slower in training that was 4.789 sec. Therefore, KNN Classification gives better performance than naïve bayes and random tree forest classification.