Cognitive load detection using hyper-learning dragonfly optimization for feature selection

Introduction Cognitive load detection is essential for understanding brain responses to mental effort and cognitive demands. Electroencephalogram (EEG) signals provide valuable information about neural activity and offer a non-invasive approach to identifying cognitive load. However, accurate detection and the identification of brain regions with superior discriminative capabilities remain important challenges. Methods This study proposes an EEG-based cognitive load detection framework using empirical Fourier decomposition (EPFD), entropy-based feature extraction, and the hyper-learning binary dragonfly feature selection algorithm (HLBDA). EEG signals recorded from each brain lobe over four-second intervals were decomposed into 10 intrinsic mode functions (IMFs). Entropy-based features were extracted from the IMFs, and the most informative features were selected using HLBDA. The selected features were classified using machine learning (ML) classifiers. The proposed framework was evaluated on the MAT, STEW, and MLR datasets. Additionally, lobe-wise cognitive load detection was performed on the MAT and STEW datasets to investigate the discriminative capabilities of five individual brain lobes. Subject-wise leave-one-out cross-validation (LOOCV) was employed to prevent data leakage. Results The proposed framework achieved classification accuracies of 97.13%, 96.48%, and 98.2% on the MAT, STEW, and MLR datasets, respectively, using the K-nearest neighbors (KNN) classifier for MAT and STEW and the ensemble subspace KNN (ESSK) classifier for MLR. In the lobe-wise analysis, the frontal lobe demonstrated superior discriminative capability for cognitive load detection, achieving accuracies of 96.7% and 96.6% on the MAT and STEW datasets, respectively, using the ESSK classifier. Discussion The results demonstrate the effectiveness of the proposed EPFD–HLBDA-based framework for EEG-based cognitive load detection across multiple datasets. The superior performance of the frontal lobe suggests its potential as a particularly informative region for distinguishing cognitive states. Furthermore, the use of subject-wise LOOCV strengthens the evaluation by reducing the risk of data leakage. Compared with recently reported methods, the proposed framework demonstrates competitive classification performance and highlights the potential of lobe-specific EEG analysis for reliable cognitive load assessment.