1 INTRODUCTION

Research in maritime safety has indicated that 75% to

90% of marine accidents are directly associated with

human error [1]. In recent years, the maritime industry

has witnessed a range of accidents with varying

degrees of severity, including but not limited to

collisions, groundings, explosions, and man-overboard

incidents [2]. Substantiating this trend, the European

Maritime Safety Agency (EMSA) reported that 80% of

ship collisions and grounding incidents over the past

decade were traceable to human-related causes [3].

These findings highlight the urgent need to address

human cognitive and behavioral determinants as a

way to reduce human error in offshore operations.

Human factors such as skill gaps, stress accumulation,

excessive mental workloads, and motivational failures

can lead to human errors [4].

In recent years, international maritime

organizations such as the International Seafarers'

Welfare and Assistance Network (ISWAN) and the

U.S. Department of Transportation's Maritime

Administration (MARAD) have strongly advocated for

the mental health and well-being of seafarers [5],[6].

These initiatives emphasize technological innovations

and scientific assessments to enhance the work

environment and cognitive health of seafarers. The

complex marine work environment demands that crew

members possess certain non-technical abilities to

carry out tasks, including psychological factors, mental

workload, and attention, which collectively shape

Seafarer Mental Workload Assessment Using a Hybrid

Deep Learning Model

R. Jiang & S. Fan

The State Key Laboratory of Maritime Technology and Safety, Wuhan, China

Wuhan University of Technology, Wuhan, China

ABSTRACT: Human errors in maritime operations are closely linked to seafarers' mental workload; however,

traditional assessment methods lack real-time neurocognitive resolution. This study introduces a novel

psychophysiological framework that integrates electroencephalography (EEG) analysis with deep learning to

objectively quantify seafarers' mental workload during onboard operations. A high-fidelity bridge simulator was

utilized to generate critical maritime scenarios, including ship encounters, narrow channel navigation, poor

visibility, and emergency responses. High-density EEG signals were analyzed to extract spectral features

(Gamma, Beta, Alpha, Theta, Delta). A hybrid Convolutional Neural Network-Bidirectional Long Short-Term

Memory (CNN-BiLSTM) model was proposed to classify workload states of seafarers, combining Convolutional

Neural Network (CNN)-extracted frequency patterns with Bidirectional Long Short-Term Memory (Bi-LSTM)-

captured temporal dynamics, which achieves 96% accuracy. Furthermore, SHAP interpretability analysis

indicated that Theta and Alpha frequencies are key indicators in distinguishing between high and low workloads

for seafarers. These results provide a quantitative tool for cognitive assessment of seafarers in maritime training

and serve as a guideline for workload allocation in ship bridge teams for shipping companies and maritime

authorities.

http://www.transnav.eu

the International Journal

on Marine Navigation

and Safety of Sea Transportation

Volume 19

Number 1

March 2025

DOI: 10.12716/1001.19.01.07

decision-making efficacy. Mental workload is the

capacity of available resources to meet task

requirements [7]. Ship operations often require

seafarers to work around the clock, which can easily

result in a heavy workload and poor decision-making.

Moreover, adverse weather conditions or high-risk

situations at sea can further increase seafarers' mental

workload, thereby elevating the risk of accidents in the

maritime environment. Consequently, investigating

workload dynamics in complex nautical contexts is

vital for advancing intelligent monitoring systems that

enhance situational awareness and critical emergency

risk management.

Traditional workload research methods, mainly

relying on subjective evaluation, can identify some

failure modes, but it is difficult to capture the

fluctuation in seafarers' neurocognitive states during

dynamic operations. For example, seafarers' mental

workload in high-risk scenarios increases significantly,

leading to distraction and delayed decision-making [8].

Existing methods cannot quantify such load changes in

real-time, thus limiting the timeliness of accident

prevention measures.

The limitations of traditional human factors

assessment methods can be addressed using a variety

of available biosignals. Among them,

electroencephalography (EEG) offers a higher degree

of accuracy. EEG equipment utilizes electrodes applied

directly to the scalp to non-invasively monitor

electrical activity in the human brain. They have highly

accurate temporal measurements and can dynamically

reflect subtle changes in mental workload [9],[10]. EEG

recognition typically involves extracting features like

spectral power from raw EEG data and the subsequent

application of the derived features to train a classifier.

Usually, the raw EEG time-domain signal is

transformed into the frequency domain using a

transform such as the Fourier transform. Commonly

considered frequency bands include the Delta (1-3 Hz),

Theta (4-7 Hz), Alpha (8-12 Hz), Beta (13-30 Hz), and

Gamma (31-80 Hz) bands. Studies have shown that

EEG band characteristics are significantly correlated

with workload levels. The power of oscillatory brain

activity in the Theta frequency range is positively

correlated with workload levels [11], and an increase in

Theta power and a suppression of Alpha power often

accompany high-workload states [12]. Nevertheless,

existing studies lack a targeted analysis of the nautical

operating environment, and few have correlated EEG

features with specific nautical events, restricting their

practical application value in maritime safety. This

study proposes to construct an event-driven EEG

mental workload assessment framework that realizes

real-time quantification of seafarers' cognitive states

and operational risk mapping through deep-learning

models. The innovations of this study are as follows: 1)

It is the first to correlate the temporal-spatial attributes

of nautical events with EEG features to construct a

scene-specific dataset; 2) A hybrid Convolutional

Neural Network-Bidirectional Long Short-Term

Memory (CNN-BiLSTM) architecture is proposed to

synergistically optimize the frequency-domain and

temporal feature characterization; 3) The combination

of interpretable analytics and events provides

theoretical support and practical guidelines for human

factors optimization in intelligent maritime systems.

2 RELATED WORK

2.1 Human factors in maritime operation

One of the core challenges in maritime safety is to

analyze the complex causal factors of human errors

[1],[2]. Among existing methodologies, Human

Reliability Analysis (HRA) has gained prominence for

its probabilistic approach to error risk assessment

through task decomposition [13]. Current HRA

variants exhibit unique limitations in addressing the

human factors within the complex marine

environment. The Technique for Human Error Rate

Prediction (THERP) [14] focuses on task-specific error

rate prediction but neglects synergistic interactions

between personnel competence, organizational

culture, and situational stressors. The Standardized

Plant Analysis Risk-Human Reliability Analysis

(SPAR-H) [15] quantifies factors affecting performance

through standardized classification criteria, but it is

difficult to capture dynamic marine environmental

influences. The Cognitive Reliability and Error

Analysis Method (CREAM) [16] contextually models

interactions among tasks and the environment, but

lacks precision in modeling hierarchical organizational

failures. The Human Error Assessment and Reduction

Technique (HEART) [17] utilizes error-producing

conditions to derive error probabilities, but

oversimplifies the dynamics of cognitive load in

emergencies. The Human Factors Analysis and

Classification System (HFACS) [18] was initially

applied to the air transportation System. In recent

years, the hierarchical classification framework

HFACS has been introduced into the maritime field to

reveal the deeper mechanisms of accidents through the

causal chain analysis of “organizational management-

supervisory failure-precursors of unsafe behaviors”.

For example, the HFACS-MA model proposed by

Chen et al. found that the interaction between

communication failure and fatigue accumulation in

ship-collision accidents is the critical path leading to

operational errors [19]. Soner et al. enhanced HFACS

with Fuzzy Cognitive Mapping (FCM) to analyze fire

hazards, revealing latent interdependencies between

technical failures and team coordination lapses [20].

Meanwhile, virtual maritime simulators have become

an important tool for human factors research, which

can synchronously collect seafarers' physiological

signals and behavioral data in controlled environments

through high-precision scenario simulation

[10],[21],[22]. Liu et al. demonstrated that simulator

training can improve seafarers' decision-making

efficiency in emergency collision avoidance tasks.

However, the assessment still relies on subjective

performance scores and lacks objective means of

quantifying cognitive load [10]. The limitations of

traditional methods have driven the exploration of

neurophysiological indicators. There is an urgent need

to integrate objective data, such as EEG signals, to

construct an objective and real-time human factors

assessment system.

2.2 EEG signals and mental workload

Recent advancements in sensor technologies have

empowered researchers to leverage physiological

metrics for investigating cognitive dimensions of

human behavior, particularly mental workload

dynamics [23]. Among biosignal modalities, EEG

stands out for its millisecond-level temporal precision,

non-invasive nature, and validated efficacy in

discriminating psychophysiological states, including

emotional arousal, cognitive load stratification, and

stress patterns [9],[24],[25],[26]. Within this context,

mental workload is operationally defined as the

cognitive resource allocation required to achieve task-

specific goals [27]. To some extent, an increase in

workload causes fluctuations in mental workload,

which has long been recognized as being strongly

associated with human performance. Therefore, it is

essential to objectively construct a workload

quantification framework.

The association of EEG, a direct physiological

representation of neural activity, and its frequency

band power characteristics with mental load has been

widely validated. The study showed that Alpha and

Theta band activity showed higher sensitivity when

test subjects increased three task load conditions,

associated with escalating acute cognitive demands

[28]. All bands are coordinated to amplify in

emergency and safety hazard situations [29]. During

the extreme fatigue phase caused by increased

workload, the local Alpha increases, and the Beta

decreases, signaling performance degradation [30].

The peak of Theta-wave power in the frontal region

fluctuates significantly with the increased workload,

reflecting increased cognitive resource depletion and

stress [31]. In conclusion, EEG waveform analysis and

its in-band power modulation can be used to measure

changes in subjects' workload during a maritime

transportation simulation task. Notably, Hogervorst et

al. benchmarked EEG against ECG and eye-tracking in

N-back task paradigms, establishing EEG’s superior

sensitivity to workload-physiology correlations [32].

Christian et al. compared subjects cross-validated with

EEG data for a multi-attribute task and 11

computational methods; the results showed that

multiple frequencies were required for EEG-based load

categorization, and compared with 10 other cognitive

state assessment methods, EEG data had a frequency

bandwidth advantage [24]. These studies further

demonstrate the reliability of EEG-based brain state

recognition in assessing subjects' mental workload

levels.

3 METHODOLOGY

3.1 Experimental design and data collection

The EEG data in this study were obtained from

database in Fan et al. [33], which constructed a multi-

scenario sailing task to simulate a real sailing

environment, as shown in Figure 1. Seafarers must

navigate a specific voyage while encountering various

events. The key events include ship encounters

(simulating multiple ship interactions), emergencies

(simulating sudden rudder failures), and poor

visibility conditions. All subjects underwent

neurological screening to exclude pre-existing

conditions and provided written informed consent

before participation. The occurrence times, types of

events, and operational responses of the participants

were recorded. To objectively define mental workload

states, EEG data were segmented into two categories:

event-driven phases and baseline phases, steady-state

navigation intervals without external disturbances,

characterized by routine monitoring tasks. This

segmentation aligns with prior studies linking

dynamic operational demands to neurophysiological

responses [10],[29]. 11 male crew members with

seafarers' certificates of competency were recruited for

the experiment (mean age 31.9 years, mean seafaring

experience 7.7 years), with the 30-minute test data from

participant 9 selected for analysis. All participants had

no history of neurological disorders and signed an

informed consent form before the experiment. The

experiments utilized a NeuroSky Mindwave single-

channel wireless EEG device (sampling rate 512 Hz) to

acquire EEG signals from the subjects. The device

employed dry-electrode technology to eliminate the

interference of conductive gel, ensuring the

participants' freedom of movement during dynamic

manipulation. To validate the performance of the

hybrid model in typical scenarios, the 30-minute data

from participant 9 was selected as the main analysis in

this study, as it covered the full range of predefined

scenarios (ship encounters, poor visibility, and

emergencies) and had the best signal quality. Future

work will extend to the full dataset to further improve

model robustness.

Figure 1. Experimental scenario[33]

3.2 Feature extraction and preprocessing

Time-frequency decomposition was performed via an

8-level Discrete Wavelet Transform (Daubechies db8

basis), segmenting raw EEG signals into five relevant

bands: Gamma (40-100 Hz), Beta (12-40 Hz), Alpha (8-

12 Hz), Theta (4-8 Hz), and Delta (0-4 Hz). Figure 2

shows the 2-second primary EEG data for test subject

9. These five EEG waves are associated with different

mental concepts. Gamma-band power reflects anxiety

under high-cognitive load, the Beta band is related to

active decision-making behavior, the Alpha band

characterizes relaxation or distraction, the Theta band

indicates emotional fluctuations, and the Delta band is

associated with fatigue or deep relaxation. To associate

EEG characteristics with nautical events, the time

window is dynamically divided based on the type and

duration of events. For the vessel-encounter event, 3-

minute windows (1.5 min pre/post-event) capture

collision-avoidance decision-making. For the poor-

visibility event, continuous 3-minute windows during

events are used for persistent strategy analysis. For the

rudder-failure event, 1-minute windows (30 s pre/post-

event) to isolate emergency response dynamics.

Longer windows addressed complex behavioral chains

in ship encounters/poor visibility, while short

windows emphasized abrupt cognitive shifts during

failures. Baseline windows represented low-demand

navigation. A sliding-window approach (512-sample

length, 32-sample step) generated a 26,393×5 temporal

feature matrix, ensuring millisecond-level EEG-event

synchronization.

Figure 2. Two-second EEG data of subject 9 was collected by

NeuroSky Mindwave headset.

3.3 Model design

The mental workload was classified into high and low

states using a hybrid CNN-BiLSTM network. The

model processes a multi-band EEG power matrix

(Gamma, Beta, Alpha, Theta, Delta) combined with

temporal dynamics through three core modules. The

first module is the CNN layer, which processes the

input data in parallel using two sets of one-

dimensional convolutional kernels sized 128 and 256,

respectively, aiming to capture short- and long-range

frequency-domain patterns through different window

sizes. After each set of convolutional operations, max-

pooling (stride=3) is utilized to compress the feature

dimensions and retain key frequency-domain

information, effectively extracting the fluctuating

features of different time scales in the EEG signals.

Following the frequency domain feature extraction in

the CNN layer, the data enters the Bi-LSTM layer. A

bidirectional LSTM layer with 64 hidden units captures

forward and backward cognitive state transitions,

augmented by dropout regularization (rate=0.4) and

layer normalization to ensure cross-scenario

robustness. Fully connected layers splice CNN

frequency-domain and LSTM temporal features. Then,

a 128-node Swish activation dense layer enables

higher-order feature interaction. Finally, the binary

classification probability of mental workload

(high/low) is output by the Sigmoid function, and the

decision boundary is optimized via cross-entropy

minimization.

4 RESULTS

4.1 Model evaluation

The CNN-BiLSTM hybrid neural network constructed

in this study demonstrates excellent performance in

mental workload assessment for seafarers. Through

the simulation data of typical scenarios such as ship

encounters, poor visibility, and rudder malfunction,

the model achieves a 96% classification accuracy in

high and low workload assessments. The area under

the curve (AUC) of the operating characteristics of the

subjects reaches 0.986, and the average precision of the

precision-recall (PR) curve is 0.99, which indicates that

the model has a remarkable ability to differentiate

between seafarers’ mental workload levels, as shown

in Figure 3.

Figure 3. ROC and PR curves of the CNN-BiLSTM model for

seafarer mental workload assessment

The confusion matrix shows that the correct

recognition rate (recall) of the high load state is 0.941.

The correct recognition rate of the low load state is

0.944, with only a few misclassifications, effectively

avoiding the omission detection problem, especially in

high-risk scenarios, as shown in Figure 4. Notably, the

model exhibited superior sensitivity to abrupt

workload transitions during emergency scenarios (e.g.,

rudder failure detection rate exceeding 95%),

validating its capability to identify event-induced

mental workload states against steady navigation

baselines.

Figure 4. Confusion matrix of high-and low-load forecasts in

the CNN-BiLSTM model

4.2 Interpretability analysis

To reveal the decision logic of the model, the Shapley

Additive Explanations (SHAP) framework is utilized

to quantitatively analyze the contribution of each EEG

band. As illustrated in Figure 5, the effects of different

bands on the classification results vary, with the Theta

band (mean |SHAP value| = 0.4346) and the Alpha

band (mean |SHAP value| = 0.3609) having the most

notable contributions. The positive effect of Theta band

indicated that the Theta oscillations in the prefrontal

region are enhanced with the increase of mental

workload, reflecting the increase in the depletion of

cognitive resources and the rise of emotional stress; the

negative effect of Alpha band corresponds to the

inhibition of attentional resources; the dynamic

balance of the two constitutes the core of the

neurocognitive load. This result indicates that the

Theta band, which reflects the emotional fluctuation

state, and the Alpha band, which represents the degree

of attentional inhibition, serve as core physiological

indicators influencing changes in seafarers' operational

performance, thereby providing a key neurocognitive

basis for the model output.

Figure 5. SHAP feature importance analysis of EEG bands in

the hybrid CNN-BiLSTM model

5 CONCLUSIONS

This study proposes an event-driven EEG load

assessment framework to objectively quantify

cognitive states in complex operational scenarios by

dynamically correlating nautical events with EEG

features. The effectiveness of the CNN-BiLSTM model

is demonstrated by its high classification accuracy and

feature fusion mechanism, which captures neural

dynamics that are difficult to resolve using traditional

methods.

The dynamic fluctuations of Theta and Alpha

oscillations provide novel neurophysiological insights

for maritime human factors research. It reflects both

short-term mental workload (Theta power surge) and

long-term attention maintenance (Alpha power

suppression), can serve as a bridge between neural

mechanisms and operational performance. In the

future, multimodal data can be further integrated to

construct a multidimensional physiological indicator

fusion model to enhance the robustness of load

assessment in complex scenarios. This study tested the

mental workload of seafarers in a more ideal state

within a simulated environment and did not consider

the factor of fatigue after long shifts. Future research

could assess the effect of fatigue accumulation on

mental workload within the context of actual seafaring

tasks. Additionally, all study subjects were male,

necessitating follow-up studies that include female

crew members to validate the generalizability of the

model. Expanding the sample size and including

seafarers with varying levels of experience is also

crucial to validating the model's ability to generalize

across individual scenarios, making it an important

direction for subsequent studies.

ACKNOWLEDGEMENT

The herein study was supported by Wuhan University

of Technology’s grant 104972025RSCrc0004.

REFERENCES

[1] R. Hanzu-Pazara et al., “Reducing of maritime accidents

caused by human factors using simulators in training

process,” J. Mar. Res., vol. 5, no. 1, Apr. 2008.

[2] V. Ronca et al., “Neurophysiological Assessment of An

Innovative Maritime Safety System in Terms of Ship

Operators’ Mental Workload, Stress, and Attention in the

Full Mission Bridge Simulator,” Brain Sci., vol. 13, no. 9,

Sep. 2023. doi: 10.3390/brainsci13091319.

[3] European Maritime Safety Agency (EMSA), “EMSA Facts

& Figures 2022,” 2022. [Online]. Available:

https://www.emsa.europa.eu/... (Accessed: Apr. 9, 2025).

[4] R. P. E. Gordon, “The contribution of human factors to

accidents in the offshore oil industry,” Reliab. Eng. Syst.

Saf., vol. 61, no. 1, pp. 95–108, Jul. 1998.

[5] International Seafarers’ Welfare and Assistance Network

(ISWAN), "ISWAN launches allyship guidance to build

safer cultures at sea," 2025. [Online]. Available:

https://www.iswan.org.uk/news/iswan-launches-

allyship-guidance-to-build-safer-cultures-at-sea/

(Accessed: May 6, 2025).

[6] U.S. Department of Transportation, Maritime

Administration (MARAD), "Every Mariner Builds A

Respectful Culture (EMBARC) Standards Program," 2025.

[Online]. Available:

https://www.maritime.dot.gov/embarc (Accessed: May 6,

2025).

[7] M. S. Young et al., “State of science: mental workload in

ergonomics,” ERGONOMICS, vol. 58, no. 1, pp. 1–17,

2015.

[8] M. Arenius, G. Athanassiou, and O. Sträter, “Systemic

assessment of the effect of mental stress and strain on

performance in a maritime ship-handling simulator,”

IFAC Proc. Vol., vol. 43, no. 13, pp. 43–46, Jan. 2010.

[9] Y. Liu et al., “EEG-based Mental Workload and Stress

Recognition of Crew Members in Maritime Virtual

Simulator: A Case Study,” in Proc. 2017 Int. Conf.

Cyberworlds (CW), Sep. 2017, pp. 64–71.

[10] Y. Liu et al., “Psychophysiological evaluation of

seafarers to improve training in maritime virtual

simulator,” Advanced Engineering Informatics, vol. 44, p.

101048, Apr. 2020.

[11] A. Gevins et al., “Monitoring working memory load

during computer-based tasks with EEG pattern

recognition methods,” Hum. Factors, vol. 40, no. 1, pp.

79–91, Mar. 1998.

[12] D. Dasari, G. Shou, and L. Ding, “ICA-Derived EEG

Correlates to Mental Fatigue, Effort, and Workload in a

Realistically Simulated Air Traffic Control Task,”Front.

Neurosci., vol. 11, p. 297, 2017.

[13] B. Kirwan, A Guide to Practical Human Reliability

Assessment, 1st ed. Boca Raton, FL: CRC Press, 1994.

[14] A. D. Swain and H. E. Guttmann, “Handbook of human-

reliability analysis with emphasis on nuclear power plant

applications. Final report,” Sandia National Lab. (SNL-

NM), Albuquerque, NM, USA, Tech. Rep. NUREG/CR-

1278, SAND-80-0200, Aug. 1983.

[15] D. I. Gertman et al., “The SPAR H human reliability

analysis method,” in Proc. 4th Int. Top. Meet. Nuclear

Plant Instrum., Control, Human-Machine Interface

Technol., pp. 17–24, 2004. [Online]. Available:

http://www.scopus.com/... (Accessed: Apr. 17, 2025).

[16] A. Modares et al., “An integrated Cognitive Reliability

and Error Analysis Method (CREAM) and optimization

for enhancing human reliability in blockchain,” Decis.

Anal., vol. 12, p. 100506, Sep. 2024.

[17] J. C. Williams, “Heart—A Proposed Method for

Achieving High Reliability in Process Operation by

Means of Human Factors Engineering Technology,”

Safety and Reliability, vol. 35, no. 3, pp. 5–25, Dec. 2015.

[18] M. Celik and S. Cebi, “Analytical HFACS for

investigating human errors in shipping accidents,” Accid.

Anal. Prev., vol. 41, no. 1, pp. 66–75, Jan. 2009.

[19] S.-T. Chen et al., “A Human and Organisational Factors

(HOFs) analysis method for marine casualties using

HFACS-Maritime Accidents (HFACS-MA),” Saf.Sci., vol.

60, pp. 105–114, Dec. 2013.

[20] O. Soner, U. Asan, and M. Celik, “Use of HFACS–FCM

in fire prevention modelling on board ships,” Saf. Sci.,

vol. 77, pp. 25–41, Aug. 2015.

[21] Y. Liu et al., “EEG-based Cadets Training and

Performance Assessment System in Maritime Virtual

Simulator,” in Proc. 2018 Int. Conf. Cyberworlds (CW),

Oct. 2018, pp. 214–220.

[22] Y. Liu et al., “Human Factor Study for Maritime

Simulator-Based Assessment of Cadets,” in Proc. ASME

35th Int. Conf. Ocean, Offshore Arctic Eng., vol. 3, New

York: ASME, 2016, p. V003T02A073.

[23] M. U. Iqbal et al., “Multi-class classification of control

room operators’ cognitive workload using the fusion of

eye-tracking and electroencephalography,” Comput.

Chem. Eng., vol. 181, p. 108526, Feb. 2024.

[24] C. A. Kothe and S. Makeig, “Estimation of task workload

from EEG data: new and current tools and perspectives,”

in Proc. Annu. Int. Conf. IEEE Eng. Med. Biol. Soc.

(EMBC), 2011, pp. 6547–6551.

[25] X. Hou et al., “CogniMeter: EEG-based Emotion, Mental

Workload and Stress Visual Monitoring,” in Proc. 2015

Int. Conf. Cyberworlds (CW), Oct. 2015, pp. 153–160.

[26] S. Shao et al., “EEG-Based Mental Workload

Classification Method Based on Hybrid Deep Learning

Model Under IoT,” IEEE J. Biomed.Health Inform., vol.

28, no. 5, pp. 2536–2546, May 2024.

[27] C. Berka et al., “EEG correlates of task engagement and

mental workload in vigilance, learning, and memory

tasks,” Aviat. Space Environ. Med., vol. 78, no. 5, Spec.

Iss., pp. B231–B244, May 2007.

[28] S. Lei and M. Roetting, “Influence of task combination on

EEG spectrum modulation for driver workload

estimation,” Hum. Factors, vol. 53, no. 2, pp. 168–179,

Apr. 2011.

[29] X. Li, L. Yang, and X. Yan, “An exploratory study of

drivers’ EEG response during emergent collision

avoidance,” J. Saf. Res., vol. 82, pp. 241–250, Sep. 2022.

[30] B. T. Jap et al., “Using EEG spectral components to assess

algorithms for detecting fatigue,” Expert Syst. Appl., vol.

36, no. 2, pp. 2352–2359, Mar. 2009.

[31] G. Shou and L. Ding, “Frontal theta EEG dynamics in a

real-world air traffic control task,” in Proc. 35th Annu. Int.

Conf. IEEE Eng. Med. Biol. Soc. (EMBC), Jul. 2013, pp.

5594–5597.

[32] M. A. Hogervorst et al., “Combining and comparing

EEG, peripheral physiology and eye-related measures for

the assessment of mental workload,” Front. Neurosci.,

vol. 8, p. 322, 2014.

[33] S. Fan et al., “Effects of seafarers’ emotion on human

performance using bridge simulation,” Ocean Eng., vol.

170, pp. 111–119, Dec. 2018.