A Comparison of Machine Learning Techniques for the Detection of Type-4 PhotoParoxysmal Responses in Electroencephalographic Signals

Photosensitivity is a neurological disorder in which the patients' brain produces different types of abnormal electrical responses, known as Photoparoxysmal Responses (PPR), to specific visual stimuli, potentially triggering an epileptic seizure in extreme cases. The diagnosis of this condition is based on the manual analysis and detection of these discharges in their electroencephalogram. This research focuses on comparing different Machine Learning techniques for the automatic detection of Type-4 PPR (the most extreme PPR) in a real EEG dataset, after the transformation using Principal Component Analysis. Different two-class and one-class classifiers are tested, and the best performing methods for Type-4 PPR detection are 2C-KNN and DL-NN. Obtained results are compared with those achieved from a previous research, resulting in a performance increase of 15%. This system is currently in study with subjects at Burgos University Hospital, Spain.