Photosensitivity is a neurophysiological condition in which the brain produces epileptic discharges, known as Photoparoxysmal Responses, as a reaction to light flashes which may lead to seizures. The standardized diagnosis procedure consists of stimulating the patient with a flashing light while Electroencephalography records their brain activity until one of these epileptic reactions occurs, which causes a scarcity of such phenomena in the recordings that, added to the low prevalence of this condition, generates an extremely imbalanced dataset which decreases the performance of automatic detection using Deep Learning techniques. This research proposes applying two different Data Augmentation techniques in the Data Generator phase of a Convolutional Neural Network to create synthetic paroxysmal instances and balance each input batch independently. Various network architectures are also tested. Results obtained from these experiments concluded that both DA techniques are equivalent in performance, but could not outperform previous experiments. This project is being developed with patients from Burgos University Hospital, Spain.