Detection of single-trial EEG of the neural correlates of familiar faces recognition using machine-learning algorithms

Abstract

We analyze Electroencephalograph EEG data with several classification algorithms to classify probe and irrelevant data. Out of eight algorithms, five foun to perform poorly: Decision Tree, Random Forest, Neural Network, SVM RBF and Adaboost, while KNN, SVM Linear and Naive Bayes Gaussian yielded satisfactorily. Analysis is carried with 14 different subjects. Various metrics like accuracy, precision and recall are calculated to establish best performing algorithms with Electroencephalogram EEG data. Further work is needed on this area by increasing the number of subjects and experiments, with an idea to eliminate inters subjective variability. Also, work on algorithms tuning for better mental states capturing.