Cognitive Workload Recognition Using EEG Signals and Machine Learning: A Review

IEEE Transactions on Cognitive and Developmental Systems, Journal Year: 2021, Volume and Issue: 14(3), P. 799 - 818

Published: June 17, 2021

Machine learning and its subfield deep techniques provide opportunities for the development of operator mental state monitoring, especially cognitive workload recognition using electroencephalogram (EEG) signals. Although a variety machine methods have been proposed recognizing via EEG recently, there does not yet exist review that covers in-depth application methods. To alleviate this gap, in article, we survey literature to identify approaches highlight primary advances. be specific, first introduce concepts learning. Then, discuss steps classical from following aspects, i.e., data preprocessing, feature extraction selection, classification method, evaluation Further, commonly used domain. Finally, expound on open problem future outlooks.

Language: Английский

---

EEG-based mental workload estimation using deep BLSTM-LSTM network and evolutionary algorithm

Biomedical Signal Processing and Control, Journal Year: 2020, Volume and Issue: 60, P. 101989 - 101989

Published: May 19, 2020

Language: Английский

---

Enhanced Accuracy for Multiclass Mental Workload Detection Using Long Short-Term Memory for Brain–Computer Interface

Frontiers in Neuroscience, Journal Year: 2020, Volume and Issue: 14

Published: June 23, 2020

Cognitive workload is one of the widely invoked human factor in areas Human Machine Interaction (HMI) and Neuroergonomics. The precise assessment cognitive mental (MWL) vital requires accurate neuroimaging to monitor evaluate states brain. In this study, we have decoded four classes using long-short term memory (LSTM) with 89.31% average accuracy for brain-Computer Interface (BCI). brain activity signals are acquired functional Near-Infrared Spectroscopy (fNIRS) from prefrontal cortex (PFC) region We performed a supervised MWL experimentation varying levels on 15 participants (both male female) 10 trials each per participant. Real-time four-level assessed fNIRS system initial classification three strong machine learning (ML) techniques, Support Vector (SVM), k-Nearest Neighbor (k-NN) Artificial Neural Network (ANN) obtained accuracies 54.33%, 54.31%, 69.36% respectively. novel Deep (DL) frameworks proposed which utilizes Convolutional (CNN) Long Short-Term Memory 87.45% respectively, solve high-dimensional problem. Statistical analysis, t- test one-way F-test (ANOVA) also through deep algorithms. Results show that DL (LSTM CNN) algorithms significantly improve performance as compared ML (SVM, ANN, k-NN)

Language: Английский

---

Measuring Mental Workload Variations in Office Work Tasks using fNIRS

International Journal of Human-Computer Studies, Journal Year: 2020, Volume and Issue: 147, P. 102580 - 102580

Published: Dec. 25, 2020

The motivation behind using physiological measures to estimate cognitive activity is typically build technology that can help people understand themselves and their work, or indeed for systems do so adapt. While functional Near Infrared Spectroscopy (fNIRS) has been shown reliably reflect manipulations of mental workload in different work tasks, we still need establish whether fNIRS differentiate variety within common office-like tasks order broaden our understanding the factors involved tracking them real working conditions. 20 healthy participants (8 females, 12 males), whose included took part a user study investigated a) sensitivity measuring variations representations everyday reading writing b) how natural interruptions are reflected data. Results supported PFC activation differentiating between levels but not terms increased oxygenated haemoglobin (O2Hb) decreased deoxygenated (HHb), harder conditions compared easier There was considerable support detecting changes due interruptions. Variations during could be understood relation spare capacity models. These findings may guide future into sustained monitoring real-world settings.

Language: Английский

---

Analyzing Classification Performance of fNIRS-BCI for Gait Rehabilitation Using Deep Neural Networks

Sensors, Journal Year: 2022, Volume and Issue: 22(5), P. 1932 - 1932

Published: March 1, 2022

This research presents a brain-computer interface (BCI) framework for brain signal classification using deep learning (DL) and machine (ML) approaches on functional near-infrared spectroscopy (fNIRS) signals. fNIRS signals of motor execution walking rest tasks are acquired from the primary cortex in brain's left hemisphere nine subjects. DL algorithms, including convolutional neural networks (CNNs), long short-term memory (LSTM), bidirectional LSTM (Bi-LSTM) used to achieve average accuracies 88.50%, 84.24%, 85.13%, respectively. For comparison purposes, three conventional ML support vector (SVM), k-nearest neighbor (k-NN), linear discriminant analysis (LDA) also classification, resulting 73.91%, 74.24%, 65.85%, study successfully demonstrates that enhanced performance fNIRS-BCI can be achieved terms accuracy compared approaches. Furthermore, control commands generated by these classifiers initiate stop gait cycle lower limb exoskeleton rehabilitation.

Language: Английский

---

Hybrid EEG-fNIRS brain-computer interface based on the non-linear features extraction and stacking ensemble learning

Journal of Applied Biomedicine, Journal Year: 2023, Volume and Issue: 43(2), P. 463 - 475

Published: April 1, 2023

The Brain-computer interface (BCI) is used to enhance the human capabilities. hybrid-BCI (hBCI) a novel concept for subtly hybridizing multiple monitoring schemes maximize advantages of each while minimizing drawbacks individual methods. Recently, researchers have started focusing on Electroencephalogram (EEG) and "Functional Near-Infrared Spectroscopy" (fNIRS) based hBCI. main reason due development artificial intelligence (AI) algorithms such as machine learning approaches better process brain signals. An original EEG-fNIRS hBCI system devised by using non-linear features mining ensemble (EL) approach. We first diminish noise artifacts from input signals digital filtering. After that, we use mining. These are "Fractal Dimension" (FD), "Higher Order Spectra" (HOS), "Recurrence Quantification Analysis" (RQA) features, Entropy features. Onward, Genetic Algorithm (GA) employed Features Selection (FS). Lastly, employ Machine Learning (ML) technique several namely, "Naïve Bayes" (NB), "Support Vector Machine" (SVM), "Random Forest" (RF), "K-Nearest Neighbor" (KNN). classifiers combined an recognizing intended activities. applicability tested publicly available multi-subject multiclass dataset. Our method has reached highest accuracy, F1-score, sensitivity 95.48%, 97.67% 97.83% respectively.

Language: Английский

---

EEG-Based Emotion Recognition Using Logistic Regression with Gaussian Kernel and Laplacian Prior and Investigation of Critical Frequency Bands

Applied Sciences, Journal Year: 2020, Volume and Issue: 10(5), P. 1619 - 1619

Published: Feb. 29, 2020

Emotion plays a nuclear part in human attention, decision-making, and communication. Electroencephalogram (EEG)-based emotion recognition has developed lot due to the application of Brain-Computer Interface (BCI) its effectiveness compared body expressions other physiological signals. Despite significant progress affective computing, is still an unexplored problem. This paper introduced Logistic Regression (LR) with Gaussian kernel Laplacian prior for EEG-based recognition. The enhances EEG data separability transformed space. promotes sparsity learned LR regressors avoid over-specification. are optimized using logistic regression via variable splitting augmented Lagrangian (LORSAL) algorithm. For simplicity, method noted as LORSAL. Experiments were conducted on dataset analysis EEG, video signals (DEAP). Various spectral features by combining electrodes (power density (PSD), differential entropy (DE), asymmetry (DASM), rational (RASM), caudality (DCAU)) extracted from different frequency bands (Delta, Theta, Alpha, Beta, Gamma, Total) Naive Bayes (NB), support vector machine (SVM), linear L1-regularization (LR_L1), L2-regularization (LR_L2) used comparison binary classification valence arousal. LORSAL obtained best accuracies (77.17% 77.03% arousal, respectively) DE total bands. also investigates critical experimental results showed superiority Gamma Beta classifying emotions. It was presented that most informative DASM DCAU had lower computational complexity relatively ideal accuracies. An recently deep learning (DL) methods included discussion. Conclusions future work final section.

Language: Английский

---

Conditional-GAN Based Data Augmentation for Deep Learning Task Classifier Improvement Using fNIRS Data

Frontiers in Big Data, Journal Year: 2021, Volume and Issue: 4

Published: July 29, 2021

Functional near-infrared spectroscopy (fNIRS) is a neuroimaging technique used for mapping the functioning human cortex. fNIRS can be widely in population studies due to technology’s economic, non-invasive, and portable nature. task classification, crucial part of with Brain-Computer Interfaces (BCIs). data are multidimensional complex, making them ideal deep learning algorithms classification. Deep Learning classifiers typically need large amount appropriately trained without over-fitting. Generative networks such cases where substantial required. Still, collection complex various constraints. Conditional Adversarial Networks (CGAN) generate artificial samples specific category improve accuracy classifier when sample size insufficient. The proposed system uses CGAN CNN enhance through augmentation. determine whether subject’s Left Finger Tap, Right or Foot Tap based on patterns. authors obtained classification 96.67% CGAN-CNN combination.

Language: Английский

---

EEG-fNIRS-based hybrid image construction and classification using CNN-LSTM

Frontiers in Neurorobotics, Journal Year: 2022, Volume and Issue: 16

Published: Aug. 31, 2022

The constantly evolving human-machine interaction and advancement in sociotechnical systems have made it essential to analyze vital human factors such as mental workload, vigilance, fatigue, stress by monitoring brain states for optimum performance safety. Similarly, signals become paramount rehabilitation assistive purposes fields brain-computer interface (BCI) closed-loop neuromodulation neurological disorders motor disabilities. complexity, non-stationary nature, low signal-to-noise ratio of pose significant challenges researchers design robust reliable BCI accurately detect meaningful changes outside the laboratory environment. Different neuroimaging modalities are used hybrid settings enhance accuracy, increase control commands, decrease time required activity detection. Functional near-infrared spectroscopy (fNIRS) electroencephalography (EEG) measure hemodynamic electrical with a good spatial temporal resolution, respectively. However, settings, where both output BCI, their data compatibility due huge discrepancy between sampling rate number channels remains challenge real-time applications. Traditional methods, downsampling channel selection, result important information loss while making compatible. In this study, we present novel recurrence plot (RP)-based time-distributed convolutional neural network long short-term memory (CNN-LSTM) algorithm integrated classification fNIRS EEG acquired first projected into non-linear dimension RPs fed CNN extract features without performing any downsampling. Then, LSTM is learn chronological time-dependence relation activity. average accuracies achieved proposed model were 78.44% fNIRS, 86.24% EEG, 88.41% EEG-fNIRS BCI. Moreover, maximum 85.9, 88.1, 92.4%, results confirm viability RP-based deep-learning successful systems.

Language: Английский

---

Cognitive workload estimation using physiological measures: a review

Cognitive Neurodynamics, Journal Year: 2023, Volume and Issue: 18(4), P. 1445 - 1465

Published: Dec. 26, 2023

Language: Английский

---