Performance Analysis of Supervised Machine Learning Algorithms Applied for Epileptic Seizures Detection

• Main Authors: Swetha, Annangi, Sinha, Arun Kumar, Jayakody, Dushantha Nalin K., Sharma, Abhishek
• Format: Conference Proceeding
• Language: English
• Subjects: Classification algorithms; Confusion matrix; Electroencephalography; Epileptic seizures; k-nearest neighbor; Machine learning; Machine learning algorithms; Multi class support vector machine; Naïve bayes; Performance analysis; Radio frequency; Random forest; Sensitivity and specificity; Software algorithms; Support vector machines
• ISSN: 2151-1810
• DOI: 10.1109/ICIAfS52090.2021.9605977

In this work, performance of four supervised machine learning algorithms namely: Multi Class Support Vector Machine (MC-SVM), K-Nearest Neighbor (KNN), Naïve Bayes (NB), and Random Forest (RF) have been evaluated for the seizure classification when patient dataset is large. The named algorithms have been used to classify the epileptic seizures into: normal seizure, partial seizures, and generalized seizure. The classification is performed on the dataset of 26,28,33 and 40 patients. The algorithms were implemented in MATLAB® R2019a software tool. Using confusion matrix the performance metrics (i.e., accuracy, precision, sensitivity and specificity) have been calculated. It is observed that the performance of the RF algorithm is best among other algorithms, when applied to the large patient datasets. For 40 patient data, the RF algorithm gives accuracy, precision, sensitivity, and specificity as 98.01 %, 95.92 %, 95.52 %, and 98.75 %, respectively, which is highest among the named algorithms.