Trust is a critical factor in effective human–automation interaction, influencing user reliance, acceptance, and system performance. As automated systems become increasingly complex and computationally advanced, real-time trust quantification is essential. This review systematically examines the literature using a PRISMA-guided approach, integrating studies on neural mechanisms, physiological measures, and computational modelling. Relevant studies were collected from multiple databases, screened according to inclusion criteria, and categorised by experimental paradigms, neural modalities, preprocessing pipelines, feature types, and machine/deep learning frameworks. We critically evaluate classical machine learning and deep learning approaches, highlighting consistent neural correlates, including oscillatory activity, connectivity patterns, and frontal and temporoparietal activations. The review also reveals a critical misalignment between psychological theories of trust, how trust is labelled in experiments, and how computational models are trained, raising concerns about the specificity and interpretability of many reported trust markers. Key challenges include limited feature exploration, dataset scarcity, real-time assessment strategies, and underutilisation of multimodal fusion across feature-, decision-, and network-level representations. By synthesising current evidence and identifying key gaps, this work provides a roadmap for developing adaptive, interpretable, and generalisable trust models that support safer and more reliable human–automation systems in complex, real-world scenarios.

Trust plays a pivotal role in shaping driver interactions with autonomous vehicles (AVs), particularly in Level 3 systems that require timely human intervention during takeover requests (TORs).While prior studies have examined trust and TOR performance independently, limited work has systematically explored their intersection.This review addresses that gap by investigating how trust is formed, miscalibrated, and recovered during TOR events.Key factors such as TOR timing, modality, environmental complexity, system transparency, and individual differences are analysed in relation to both trust and takeover performance.Current trust models and measurement techniques are critically evaluated, highlighting limitations of static approaches and the emerging value of real-time trust monitoring.Guided by PRISMA 2020, a systematic literature review was conducted to screen and synthesise relevant studies.The review identifies challenges and offers design recommendations for adaptive, trust-sensitive AV systems that foster calibrated trust, ultimately improving safety, driver readiness, and overall user acceptance in automated driving contexts.

In today’s world, intelligent machines play increasingly important roles in collaboration with humans, necessitating a deeper level of trust between individuals and technology. However, there is a significant research gap in understanding how humans calibrate their trust in automation, which is crucial for optimising system usage and preventing the misuse or disuse of automated systems. In this article, we introduce an artificial neural network (ANN) framework designed to classify trust and distrust in automation based on electroencephalogram (EEG) signals. Our objective is to develop an effective model that can accurately differentiate between states of trust and distrust exhibited by individuals towards automated systems. In this research, we utilised an existing online dataset specifically focused on eliciting trust and distrust states. This dataset served as the foundation for feature extraction and selection, training, and validation of the proposed ANN framework, which is aimed at robustly classifying trust and distrust based on EEG data. To evaluate the effectiveness, the performance of the proposed ANN framework was compared with three traditional machine learning methods: Naive Bayes, K-Nearest Neighbour, and Support Vector Machine. Performance metrics such as accuracy, sensitivity, and specificity were used for trust and distrust classification. The results demonstrated the efficacy of the ANN approach in leveraging EEG data to enhance the classification of trust and distrust towards automation. This research contributes to advancing the understanding of human trust dynamics in human-machine interaction, which is essential for developing more reliable and trustworthy automated systems.

Fatigue poses a significant challenge to safety and performance across various domains, making its accurate detection and classification essential for preventing accidents and enhancing operational efficiency. Fatigue classification using an electroencephalogram (EEG) plays a critical role in applications including occupational health, transportation safety, and performance monitoring in high-stakes environments. Deep learning models, including convolutional neural networks (CNNs) and recurrent neural networks (RNNs), have shown potential for analysing EEG data but often face limitations in capturing both temporal dependencies and long-range patterns critical for fatigue detection. This study investigates the use of a transformer-based temporal convolutional network (TCN-Transformer) for EEG-based fatigue classification, leveraging the TCN’s causal and dilated convolutional architecture in conjunction with transformer components to efficiently extract multi-scale temporal features and model long-range dependencies in sequential data. Comparative experiments demonstrate that the transformer-enhanced TCN framework outperforms CNN and RNN models for real-time fatigue classification, achieving an accuracy of 92.70%, a sensitivity of 92.69%, and an AUC of 0.9801. The evaluation strategy, employing 10-fold cross-validation repeated over 10 runs, validated the robustness of the proposed approach. This study establishes transformer-based TCNs as a promising tool for EEG-based fatigue classification, providing a streamlined and effective method for real-time fatigue monitoring in safety-critical environments.

The increasing sophistication of autonomous systems and robotics has spurred research into the unique dynamics of human-autonomy teaming (HAT). These advanced technologies aim to enhance decision-making, situational awareness, mutual understanding, and interpersonal relationships, thereby minimizing risks in collaborative endeavors. However, achieving effective HAT requires careful attention to team trust, particularly in scenarios that move beyond simple human–machine dyads to involve complex, multiagent systems and distributed teams. This review undertakes a comprehensive exploration and evaluation of trust measurement techniques within the domain of HAT, focusing on methods that are sensitive to the nuances of these complex team dynamics. Emphasising the significance of trust measurement in optimising team performance, the review categorizes existing empirical works into subjective (e.g., self-report, questionnaires, surveys) and objective (e.g., behavioral, physiological) indices. Drawing insights from recent literature (2019–2024), the article explores the complexities of trust measurement, addressing methodologies employed by researchers and synthesizing their findings. The study suggests directions for further investigation into improving trust assessment techniques and developing practical models to better suit the evolving context of human-autonomy collaboration. The review highlights gaps in current methods and suggests avenues for future research, particularly in refining trust calibration models for dynamic, evolving contexts in human-autonomy collaboration and for understanding how trust is distributed and managed across complex team structures.

Measuring team performance and physiological compliance (PC) within a team have gained interest in the last few decades. The team’s performance or functioning of a team is overseen by attributes such as collaboration, coordination, attitudes and motivation. Team emergent states such as situational awareness, trust, emotions and mutual understanding influence the attributes of teams. This study examines the relationship between PC and collaboration. It also investigates how the cognitive state influences this relationship. Seventeen teams, each with three members, participated in a collaborative simulated task, while their electrocardiogram (ECG) activity was recorded via a chest strap device. Short-term time-domain measures of heart rate variability (HRV) were derived for each participant. PC was established using the mean of the cross-correlation (CC) between dyads within a team. A linear regression model was employed to examine the relationship between PC and self-reported measures of team collaboration. This in turn will be applied to investigate how cognitive state between team members influences that relationship. The results have shown a statistically significant (p<0.05) positive relationship between PC and collaboration. In conclusion, PC has the potential to be an objective method to quantity teamwork effectiveness where it can be assessed via self-reported collaboration.

In this research, we present a new emotion recognition model using wavelet packet energy features using electroencephalography (EEG) data. Wavelet packets has been widely used as a means of time-frequency analysis of EEG in many different applications including brain computer interface systems. Features for emotion recognition are extracted from the EEG signals using a depth 6 wavelet packet tree. Wavelet packet energy of the sub-bands corresponding to delta (0-4Hz), theta (4-8Hz), alpha (8-13Hz), beta (13-30Hz), and gamma (30-49Hz) are taken as emotional features. Feature selection based on feature ranking is applied to select the most prominent EEG channel-frequency combinations for emotion recognition. Four classical classifiers such as linear discriminant analysis, support vector machine, K nearest neighbor and naive Bayes were used to detect the emotional states from the extracted features. To evaluate the effectiveness and validation of the proposed model, the SEED database has been employed. Based on the experiment results obtained, our method demonstrates that the LDA is more suitable for emotion recognition as compared to other classical classifier, which achieving the best average accuracy of 90.9386. Emotion recognition systems with high accuracy give opportunities to study real world applications such as mental state and fatigue monitoring.

Employing electroencephalography (EEG) data for classifying emotion has attracted a significant interest among researchers. With the ever advancing machine learning and neural network computational power, several models have been proposed for classifying the emotion states. Training an efficient classifier requires selection of appropriate features demanding complex computations. The present article investigates whether dense convolutional neural networks can extract the features from just pre-processed raw EEG signals enabling a suitability for online classification. The article presents a fully connected multi-channel neural network model comprising of dense con-volutional layers coupled with sparse autoencoder and dense perceptron layers for classifying emotions from 62-channel EEG data. The proposed model uses the dense convolution layers to learn emotion features directly from each channel. The cascaded SAE layer coupled with dense perceptron network learns correlations between the channels for efficient classification. In addition, different methods exist to generate training and testing sets for validating classifier models. In this light, the article discusses three most recent methods to generate training and testing sets employed in emotion classification. The proposed model achieves a best testing accuracy of 97.42%.

Habituation responses to startle stimuli have been linked to individuals' resilience assessment. Derived from skin conductance levels (SCL), individuals' who elicit quicker habituation to startle stimuli typically report possessing higher trait resilience. However, this link has not yet been evaluated in the context of individuals' dynamic resilience. This study examines whether the habituation of skin conductance response (SCR) could be used as a predictive measure for classifying individuals into high, moderate and low dynamic resilience groups. A dynamic decision-making task was paired with an acoustic startle paradigm, in which 53 participants were analyzed while being subjected to acoustic startle stimuli during the task. SCL of the participants were continuously recorded during the 24-minute trials. On the contrary to previous results in the literature supporting a link between skin conductance levels and trait resilience, current results did not show a solid predictive power toward quantification of dynamic resilience. However, the results act as a first step and open the way for further research to consider testing SCL measures as predictors of dynamic resilience in more ecological task environments than in controlled laboratory settings. These more ecological task environments may yield more SCL sensitivity for dynamic resilience measurement, as individuals will experience more naturally fluctuating task load and be required to activate various combinations of cognitive functions to cope with the task stressors.

Brain-computer interface (BCI) has enormous potential applications from rehabilitation of neural disease patient to driver's drowsiness prediction. However, the feasibility of BCI in the practical application other than laboratory prototype is dependent on the usability of a human subject on the go. Most of the time, it needs long and in-depth calibration of the system for training purpose. Different novel methods have been proposed to shorten the calibration time maintaining the robust performance. One of them is active learning (AL) which asks for labeling the training samples and it has the potential to reach robust performance using reduced informative training set. In this work, one of the AL methods, query by committee (QBC) is applied by three state-of-the-art feature extraction methods coupled with linear discriminant analysis classifier for motor imagery-based BCI. The joint accuracy by the members of QBC has obtained the baselines using maximum 33% of the whole training set. It also shows a significant difference at the 5% significance level from contemporary AL methods and random sampling method. Thus, QBC has reduced the labeling effort as well as the training data collection effort significantly more than that of random labeling process. It infers that QBC is a potential candidate for abridging overall calibration time of BCI systems.

An ideal noninvasive electroencephalography (EEG) based brain-computer interface (BCI) is a user-friendly plug and play system where a new user does not need to go through the long training data collection process. To reduce the amount of training data required for a new user, active learning inspired informative instance transfer is investigated in this work as one of the potential solutions. In this informative transfer learning, query by committee is applied as query method to find informative samples from subjects own domain. On the other hand, percentile range around the mean of center distance (PRMCD) query method is introduced in this work as an alternative to existing entropy criterion to find informative samples from the past user's domain. The newly introduced PRMCD algorithm has reached the benchmark performance using only average 12% of whole subjective training set while the existing entropy-based algorithm has achieved the benchmark performance using average 17% of the whole subjective training set in case of 7 out of 9 subjects. For PRMCD algorithm, a new user can achieve the intended mean benchmark performance using reduced (only 50 which is 12.5%) amount of training data in general irrespective of subjects. Therefore, incorporation of PRMCD algorithm has added an important step towards the zero training BCI. It is a significant advancement for the practical application of motor imagery based BCI.

Steady state visual evoked potential (SSVEP)-based brain computer interface (BCI) systems has attracted paramount amount of attention due to their higher signal to noise ratio and high information transfer rate. In this paper a SSVEP-BCI-based on a convolutional neural network (CNN) classifier is presented. The visual stimulation is provided to the participants with with LED matrices blinking at 6, 7, 8 and 9 Hz respectively. A wireless EEG amplifier, the g.Nautilus was used to acquire the electroencephalogram (EEG) signals from eight parietal and occipital electrodes. The features were derived using Fast Fourier Transformation (FFT) of the 8 channels using a 2s moving window in the form of 8 × 8 grey scale images. The proposed CNN architecture has provided superior average accuracy of 94.7% for four subjects, compared to the average accuracy of 87.4% of the state of the art canonical correlation analysis (CCA) performance.

Cognitive load is an often-discussed important topic with regards to human performance. Currently, many psychophysiological measures are used to quantify the level of perceived cognitive load under different tasks and environments. Heart rate (HR) is reported in literature as one of the physiological parameters that is influenced by varying cognitive load levels. Electrocardiography (ECG) is the gold-standard measure of HR measurement, however the use of traditional ECG measurement systems limits the applicability of the system to a lab environment. Recent advancements in wearable devices have provided a step towards bringing the physiological signal based human performance measuring system into real-world applications. In this study we are investigating the usability of the Polar OH1, a HR monitoring device initially used for the purpose of physical activity monitoring to use in an arithmetic cognitive load task. With a study carried out with a dataset of 10 subjects, we are able to conclude that the Polar OH1 can be used in place of ECG monitored HR, at varying cognitive load levels.

Brain-computer interface (BCI) using machine learning has the requirement for a large number of training data. This requirement makes the long training session inevitable for a new user. Many approaches including transfer learning (TL) already have been reported in the literature to abridge the long training data collection session. One of them is transferring informative instances using active learning (AL) which was approached in our previous attempts for both binary and multiclass BCI. It was associated with the classic common spatial pattern (CSP) feature extraction method. It showed the potential to obtain the benchmark performance using a reduced amount of training data. However, it has subject dependent performance and was not up to the expectation particularly for multiclass BCI. For binary BCI, it is addressed by selecting the best subject-specific features from subjective narrow frequency window using filter bank CSP (FBCSP). Since multiclass BCI has different characteristics in terms of output performance and nature of features, this work investigates the incorporation of FBCSP into informative transfer learning with AL (ITAL) for multiclass BCI. Comparing with existing direct transfer with AL (DTAL) and ITAL with CSP for multiclass BCI, ITAL with FBCSP reaches the benchmark performance for six out of nine subjects using average 42% of the full training set which is significant at 5% (p <; 0.05) significance level. For multiclass BCI as well, ITAL combined with discriminating feature extraction ensures better transfer which yields to effective reduction of the training session without sacrificing the benchmark robustness.

Multi-screens or wide screen setup is becoming increasingly popular in many work places and training environments. However, there has been limited studies of their effect on human health and performance. In this study, we investigate individuals performance and eye movement characteristics while performing a visual task on multi-screen setup consisting of three monitors. During the task, subjects had to share their attention among three screens to identify the location of the visual stimulus and respond accordingly. Subjects' score was calculated based on validity of input to stimulus and response time, while fixation characteristics were investigated with respect to eye movements. The results show that the use of 3-screens added extra demand on the individual causing a decrease on the score and decrease in reaction time. In a further investigation, we found statistically significant negative correlation between the task score and the participant's age while a statistically significant positive correlation between the response time and the participant's age. In addition, the use of multi-screens to perform the tasks caused both fixation occurrences and duration to decrease, denoting an increased alertness since respond was given with less fixation duration and occurrences and less response time.

This paper presents an approach to analysis of multiclass EEG data obtained from the brain computer interface (BCI) applications. The proposed approach comprises two stages including feature extraction using the common spatial pattern (CSP) and classification using fuzzy logic systems (FLS). CSP is used to extract significant features that are then fed into FLS as inputs for classification. The metaheuristic population-based particle swarm optimization method is used to train parameters of the FLS. The multiclass motor imagery dataset IIa from the BCI competition IV is used for experiments to highlight the superiority of the proposed approach against competing methods, which include linear discriminant analysis, naïve bayes, k-nearest neighbour, ensemble learning AdaBoost and support vector machine. Results from experiments show the great accuracy of the combination of CSP and FLS. Therefore, the proposed approach can be implemented effectively in the practical BCI systems, which would be helpful for people with impairments and rehabilitation.

Motor imagery based brain computer interface (BCI) has drawback of long subject dependent calibration session times. This can be a very exhausting and a time consuming process. In order to alleviate it, transfer learning and active learning approaches can be utilised. Informative instances are selected by applying active learning concept from other subjects under similar circumstances. Then, they are transferred to target user domain which has low number of training data. This informative transfer learning approach is associated with common spatial pattern (CSP) as feature extraction method in our previous attempt. CSP features are widely used for motor imagery-based BCI systems. However, the classical CSP algorithm will perform poorly when operational frequency bands are inadequately selected. Therefore, in the present study, filter bank common spatial pattern (FBCSP) algorithm has been applied for extracting features from the multi-class motor imagery data. FBCSP algorithm selects subject-specific operational frequency bands for extracting discriminative features. We incorporated FBCSP features into informative instance transfer learning framework to investigate the effect of subject specific feature selection. Results show that performance of new users can be improved with reduced number of training samples when FBCSP features are used compared to the classical CSP-based features.

An approach to EEG signal classification for brain-computer interface (BCI) application using fuzzy standard additive model is introduced in this paper. The Wilcoxon test is employed to rank wavelet coefficients. Top ranking wavelets are used to form a feature set that serves as inputs to the fuzzy classifiers. Experiments are carried out using two benchmark datasets, Ia and Ib, downloaded from the BCI competition II. Prevalent classifiers including feedforward neural network, support vector machine, k-nearest neighbours, ensemble learning Adaboost and adaptive neuro-fuzzy inference system are also implemented for comparisons. Experimental results show the dominance of the proposed method against competing approaches.

Event related potential (ERP) analysis is one of the most widely used methods in cognitive neuroscience research to study the physiological correlates of sensory, perceptual and cognitive activity associated with processing information. To this end information flow or dynamic effective connectivity analysis is a vital technique to understand the higher cognitive processing under different events. In this paper we present a Granger causality (GC)-based connectivity estimation applied to ERP data analysis. In contrast to the generally used strictly causal multivariate autoregressive model, we use an extended multivariate autoregressive model (eMVAR) which also accounts for any instantaneous interaction among variables under consideration. The experimental data used in the paper is based on a single subject data set for erroneous button press response from a two-back with feedback continuous performance task (CPT). In order to demonstrate the feasibility of application of eMVAR models in source space connectivity studies, we use cortical source time series data estimated using blind source separation or independent component analysis (ICA) for this data set.

Adaptive autoregressive (AAR) modeling of the EEG time series and the AAR parameters has been widely used in Brain computer interface (BCI) systems as input features for the classification stage. Multivariate adaptive autoregressive modeling (MVAAR) also has been used in literature. This paper revisits the use of MVAAR models and propose the use of adaptive Kalman filter (AKF) for estimating the MVAAR parameters as features in a motor imagery BCI application. The AKF approach is compared to the alternative short time moving window (STMW) MVAAR parameter estimation approach. Though the two MVAAR methods show a nearly equal classification accuracy, the AKF possess the advantage of higher estimation update rates making it easily adoptable for on-line BCI systems.

Understanding neural functions requires the observation of the activities of single neurons that are represented via electrophysiological data. Processing and understanding these data are challenging problems in biomedical engineering. A microelectrode commonly records the activity of multiple neurons. Spike sorting is a process of classifying every single action potential (spike) to a particular neuron. This paper proposes a combination between diffusion maps (DM) and mean shift clustering method for spike sorting. DM is utilized to extract spike features, which are highly capable of discriminating different spike shapes. Mean shift clustering provides an automatic unsupervised clustering, which takes extracted features from DM as inputs. Experimental results show a noticeable dominance of the features extracted by DM compared to those selected by wavelet transformation (WT). Accordingly, the proposed integrated method is significantly superior to the popular existing combination of WT and superparamagnetic clustering regarding spike sorting accuracy.