Detecção de epilepsia por meio de análise de eletroencefalograma utilizando aprendizado de máquina

Abstract

Epilepsy affects over 50 million people all over the world, its diagnosis is made mainly through the use of electroencephalogram. Due to the lack of qualified professionals and the complexity of the exam’s interpretation, there’s a high inaccuracy in the diagnosis of epilepsy. Many papers have proposed the use of machine learning techniques to help health professionals in their decision-making. In the present study, a classifier model was developed and tested using Bonn’s Database, which contains electroencephalograms divided into 5 groups. Wavelet Transform is used for feature extraction, and classification is done using many techniques. The following classifiers are used: Artificial Neural Networks, Support Vector Machine, K Nearest Neighbours, Gaussian Naive Bayes, Decision Tree, and Quadratic Discriminant Analysis. The results indicate similar performance in the binary classification cases, with mean accuracy of 99.63% on case B-E, and 95.85% on case D-E.In the multiclass classification case A-B-C-D-E the neural network-based technique and the decision tree resulted in better accuracy, 82.4% and 82.2%, respectively.