Characterization of memory load in an arithmetic task using non-linear analysis of EEG signals

In this paper, we investigate non-linear analysis of electroencephalogram (EEG) signals to examine changes in working memory load during the performance of a cognitive task with varying difficulty levels. EEG signals were recorded during an arithmetic task while the induced load was varying in seven levels from very easy to extremely difficult. The EEG signals were analyzed using three different non-linear/dynamic measures; namely: correlation dimension, Hurst exponent and approximate entropy. Experimental results show that the values of the measures extracted from the delta frequency band of signals acquired from the frontal and occipital lobes of the brain vary in accordance with the task difficulty level induced. The values of the correlation dimension increased as the task difficulty increased, showing a rise in complexity of the EEG signals, while the values of the Hurst exponent and approximate entropy decreased as task difficulty increased, indicating more regularity and predictability in the signals.