A Comparative Study on Classification of Working Memory Tasks Using EEG Signals

Assessing a mental workload level using electroencephalography (EEG) signals represents an active research area. The development of low-cost wireless EEG headsets drew the attention of researchers in the field of critical human-machine collaboration systems. In this paper, some classification methods are used to discriminate the working memory load levels using EEG raw data records. The brain waves were acquired during several tasks performed according to the n-back paradigm. Selecting the right features and assessing the workload memory level remains a challenging task for a broad range of practical applications. The preprocessed EEG signals are classified with different algorithms suitable for non-stationary signals. The proposed experiments guide an empirical feature selection and correlate EEG signals classification based on n-back memory tests using different acquisition devices. A professional Brain Products helmet and a low-cost wireless Emotiv Epoc+ device were used for comparison. The results prove the impact of two multiclass classifiers, Random Forests (RF) and Support Vector Machine (SVM) and feature selection, on EEG signals classification for n-back working memory tasks and highlight some disadvantages of wireless EEG acquisition tools. The results obtained suggest that both RF and SVM classifiers performed well on EEG data. Workload levels can be precisely identified using RF and a limited set of frontally located electrodes for data acquisition on both EEG devices.