EEG analysis of Evoked Potentials of the brain to develop a mathematical model for classifying Tinnitus datasets

Tinnitus is hearing a sound of buzzing, ringing, whooshing, etc. when there are no actual sounds existing specially when the background is quiet. Per statistics from the American Tinnitus Association, these symptoms affect twenty percent of population's life. In this study, we investigate 12 individuals distributed between 6 normal subjects and 6 subjects suffering from Tinnitus to develop a mathematical model for identifying Tinnitus patients in compare with normal subjects using a 14-channel low cost commodity neuroheadset (Emotiv). Our pipeline involves collecting Electroencephalography (EEG) data from the 12 subjects. We then perform noise reduction, after that we split the data into training and testing datasets, followed by labeling, fusion and randomization using Independent Component Analysis approach to then be passed to several classification algorithms to be compared and chosen from the best candidate models based on the best calculated accuracy. We compare Support Vector Machine approach versus K Nearest Neighbor as final models. We then validate the selected model using the test data resulting in a model capable of classifying EEG data as Tinnitus or not. Our method demonstrates that commodity EEG neuroheadsets can be used to identify Tinnitus patients using our proposed model.