Cognitive load detection using Ci-SSA for EEG signal decomposition and nature-inspired feature selection

Cognitive load detection is eminent during the mental assignment of neural activity because it indicates how the brain reacts to stimuli. The level of cognitive load experienced during mental arithmetic tasks can be determined using an electroencephalogram (EEG). The EEG data were collected from publicly available datasets, namely, mental arithmetic task (MAT) and simultaneous task workload (STEW). The first phase comprises decomposing the electroencephalogram (EEG) signal into intrinsic mode functions (IMFs) using circulant singular spectrum analysis (Ci-SSA). In the second phase, entropy-based features were evaluated using IMFs. After that, the extracted features were fed to nature-inspired feature selection algorithms: genetic algorithm (GA), binary particle swarm optimization (BPSO), particle swarm optimization (PSO), binary bat algorithm (BBA), and binary dragonfly algorithm (BDA) for optimal selection of features by using machine learning (ML) techniques: K-nearest neighbor (KNN), support vector machine (SVM) to analyse the classification accuracy (Ac), sensitivity (Se), specificity (Sp), precision (Pr), and F-score with 10-fold cross-validation in the third phase. The highest classification Ac, Se, Sp, Pr, and F-score of the MAT dataset were 97.30%, 0.98, 0.97, and 97.40% from multileads, and 96.20%, 0.96, 0.94, and 96.70% from a single lead (F4) of EEG, respectively. However, we achieved 97.98%, 0.98, 0.98, 0.97, and 98.1% values from multi-leads and 96.67%, 0.96, 0.97, 0.95, and 96.90% from a single-lead STEW dataset. When compared to previous state-of-the-art methods, the proposed method (Ci-SSA+BDA+KNN) has proven to be more successful.