Effects of Dimension Reduction Methods on Boosting Algorithms for Better Prediction Accuracies on Classifications of Stress EEGs

This research aims to investigate the effects of various dimension reduction methods, namely Principal Component Analysis (PCA), Independent Component Analysis (ICA) and Linear Discriminant Analysis (LDA) on the prediction accuracies of the stressful state in the EEG signaling when performing different mental tasks. The dataset used for this research is the SAM-40 dataset. It consists of 40 subjects performing three different mental tasks, i.e. Arithmetic Problem Solving Task, Stroop Color Word Test and Mirror Image Recognition Task. Each task was carried out with 3 trials. The results after applying the different dimension reduction methods of PCA, ICA and LDA to different boosting algorithms were analyzed and compared meticulously. These boosting algorithms are mainly the ensemble techniques of AdaBoostM1 and RUSBoost algorithms. Among all the experimented results shown, the LDA induced boosted classification methods showed the best prediction accuracy result, i.e. around $30 \%$ of prediction accuracy improvement.