Machine learning based Visual Evoked Potential (VEP) Signals Recognition

Introduction: Visual evoked potentials contain certain diagnostic information which have proved to be of importance in the visual systems functional integrity. Due to substantial decrease of amplitude in extra macular stimulation in commonly used pattern VEPs, differentiating normal and abnormal signals can prove to be quite an obstacle. Due to developments of use of machine learning techniques in interdisciplinary fields, deployment of machine learning in the brain electrical activity fields results in use of less expensive databases and prevention of duplication and forgery of activities.   Materials and Methods: In this study 54 normal Visual evoked potentials and 16 abnormal VEPs have been used. Signals have been classified via two main supervised learning methods, neural network and support vector machine.   Results: The results of these supervised learning techniques have been compared with similar models post feature extraction carried out by Daubechies wavelet feature extraction. Results indicate best error rate of %1.45 in SVM and %7.25 in neural network prior to feature selection via wavelet. After applying wavelet transform, SVM accuracy increased to %100 accuracy and %94.22.   Conclusion: The choice of a suitable feature selection method besides SVM and neural network can prove to be highly compatible in the field of brain electrical activity fields.