Mental Task Classification Using Deep Transfer Learning with Random Forest Classifier

Abstract

A BCI theoretical idea is to construct an output feature or task for a user using brain signals. These signals are then transmitted to the machine where the required task is performed. In this work, we present a mental task classification model that focuses on the notion of transfer learning and addresses the issues of data scarcity, choice of model selection, and low-performance measure. To decide the optimal network for feature extraction, we used five different pre-trained networks including VGG16, VGG19, ResNet101, ResNet18, and ResNet50. For the classification, the suggested model experiments with three baseline classifiers namely support vector machine, decision tree, and random forest. The model's experimental evaluation is done on the publicly available Keirn and Aunon databases. From the experiment, it is observed that features extracted from the transfer learning models help to identify the five different mental tasks efficiently. The highest average accuracy of 81.25% is attained on ResNet50 based features with a random forest classifier.