Table 3.

Brief review of the selected 32 papers.

Ref	Parameter	Type of Study	Hardware	Test Variable	Data Analysis	Findings
[54]	HR, HRV	Simulator	Chest belt HR/RR monitor (BioHarness 3, Zephyr technology)	Driving task difficulty	One-factorial repeated measures ANOVA	Increase in task difficulty caused decrease in LF-HRV, decrease in HF-HRV.
[58]	HR, HRV, RR	Simulator	Biopac MP150 (BIOPAC Hardware)	Reaction time to peripheral stress inducing driving tasks	ML: K-Nearest Neighbour, Nearest Mean Classifier, Multilayer perceptron	Machine learning classification algorithms (MLP, K-NN) classified uncorrelated features to achieve 91% accuracy.
[75]	HR, HRV	On-road	Zephyr BioHarness 3 (Zephyr technology)	Secondary task interaction with in-vehicle infotainment system	Repeated measure ANOVA	Increase in HR from single-task to dual-task condition.
[59]	HR, HRV, RR	Simulator	G-Tech medical sensors (G-Tech Medical Inc. Austria)	Traffic intensity	Pearson correlation	HRV features has high correlation with subjective feedback. Moderate correlation with breathing frequency.
[76]	HR, HRV	Simulator	Polar H7 heart monitor (Polar Electro Inc.)	Rural vs urban vs motorway scenario	Multivariate ANOVA, Post-hoc Pairwise comparison	Pairwise comparison showed RMSSD and pNN50 variance as insignificant between motorway and rural roads. VLF variance was insignificant between rest and rural scenarios.
[56]	HR, HRV, RR	Simulator	Portable Mobii ECG device, RespiV6 sensor (TMS International BV)	Time-pressure vs non time-pressure scenario	Paired t-tests	Increased HR, RR in time-pressure scenario. Task needs to be constant with time to interpret HRV features.
[60]	HR, HRV	On-road	ErgoLAB ECG sensors (Grennlee ErgoLAB)	Intersection turning behaviour	SDNN and RMSSD comparison	SDNN and RMSSD was higher in turning scenarios. No changes in pNN50 were observed.
[61]	HR, HRV, RR	Simulator	Biopac MP100 (BIOPAC Hardware)	Critical driving scenario	Multivariate ANOVA, Post-hoc analysis with univariate measures	Increase in HR in dual-task conditions. No significant changes in RR.
[62]	HR, HRV	Simulator	MEDAC ECG System (MEDAC Tech Co., Ltd.)	N-back tasks	One-factor within subject ANOVA, Artificial Neural Network classification	Time domain measures were more sensitive (IBI, RMSSD, SDNN) and decreased with increase in task difficulty.
[52]	HR, HRV	On-road	Electrocardiograph 300 G (Bionex Medical Equipemnt)	City vs rural vs motorway roads	Relationship between mean HR and frustration factor subjective data	High correlation between subjective and R-R interval. Least mean RR was observed on city roads, high mean RR was observed on motorways.
[54]	HR, HRV	On-road	Eight-slot Bionex ECG 5003711-08 (Bionex Medical Equipment)	Drive by speed signs	Repeated measures ANOVA	One cluster of participants responded strongly to intensive braking, increasing HR and decreasing mid-band HR. Task similarity affects MWL.
[79]	HR, HRV	On-road	Optical sensor (HR)—Atmel AtMega328 P (Microchip Technology)	Traffic density, effect of autopilot feature, occupant interaction	ANOVA	HR and HRV features did not show any significant effects from environmental changes or use of the autopilot feature. Subjective ratings suggest reduced workload with automation in automation-experienced drivers.
[63]	HR, HRV, RR	On-road	MLA2505 biopotential ECG, MLT1132 respiratory belt (ADInstruments)	Lane changing behaviour	No statistical tests as smaller sample size	HR decreases with time on task. The LF/HF features of HRV were affected by driver posture, vehicle vibration and did not yield the same results as observed in subjective feedback.
[64]	HR, HRV	On-road	Biopac—Bionomadix (BIOPAC Hardware)	Impact of lane keeping assisting system (LKAS)	ANOVA with repeated measures, Post-hoc pairwise comparison	No major impact of LKAS on HR, HRV. HR was higher on curvy road when compared to straight motorway.
[65]	HR, HRV	Simulator	Biopac MP36R (BIOPAC Hardware)	Types of music	Mixed ANOVA	Mean HR difference was noticed in sanguine drivers with rock music. Phlegmatic drivers showed low arousal levels in general and has high tolerance to stimulus. No significant difference for melancholic drivers. Driving impairment was observed in choleric drivers.
[82]	HR, HRV	Simulator	Android smartwatch-PPG (Manufacturer not specified)	Weather conditions, road type, passenger presence	Comparison of RMSSD	RMSSD is used as a dependent variable. HRV increases in highways compared to city roads. Stress levels decreased in the presence of a passenger.
[66]	HR, HRV	Simulator	Biopac MP150 (BIOPAC Hardware)	Effect of fatigue, gender-related differences	Mann-Whitney U tests	HRV time and frequency features domain tend to show significant difference between alert and fatigued states. HRV time and frequency domain features had gender differences in detecting mental workload.
[77]	HR, HRV	Simulator	Polar H10 heart monitor (Polar Electro Inc.)	Types of Music	Mixed-model condition x personality (M)ANOVA, Pairwise comparisons	Mean HR did not reveal any significant differences to different auditory input.
[57]	HR, HRV	Simulator	Polar H10 heart monitor (Polar Electro Inc.)	N-back tasks	One-way ANOVA	HR was found to be higher in urban and motorway scenarios. Time domain measures of HR and HRV seemed to be more sensitive to changes in MWL.
[81]	HR, HRV	On-road	Ear lobe PPG-based sensor (Manufacturer not specified)	Road geometry	ANOVA, ML: K-Means clustering	An increase in visual input increased HR. The effect of geometry on road is not very different among participants of different age, occupation, driving experience or reaction time.
[67]	HR, HRV	Simulator	PhysioLab wireless ECG device (PhysioLav Co., Ltd)	Illumination, longitudinal road slope	Two-way repeated measures ANOVA	Inconsistent results were observed in HRV time domain features due to changes in luminance and road slope.
[68]	HR, HRV	Simulator	Biopac MP160 BioNomadix (BIOPAC Hardware)	Traffic density, weather conditions, speed, highway vs city vs dual carriage roads	Parametric statistical analysis (T-scores)	Highway—increased HR and HF/LF ratio Interurban—increased HR and lower RMSSD Urban low-complexity—increased HR, LF/HF ratio, reduced RMSSD
[69]	HR, HRV	Simulator	Encephalan Mini (Medicom MTD system, Taganrog, Russia)	Short-term vs long-term verbal learning	ML: Decision Trees, Discriminant Analysis, Logistic regression, Support Vector Machines, Nearest Neighbour, Ensemble classifiers	HF/LF features of HRV showed significant changes only for verbal and auditory learning tests. The best classification accuracies were achieved with multimodal IR + HRV features.
[78]	HR, HRV	Simulator	Polar H10 heart monitor (Polar Electro Inc.)	Secondary task (In-vehicle interference)	One-sample two-sided t-tests	Significant effect of task difficulty on HR, the lower difference in HR in task conditions compared to no task conditions. Significant changes in the number of successes in task performance.
[80]	HR, HRV	Simulator	Smartwatch-PPG (Manufacturer not specified)	Traffic density, stress levels	Log-likelihood	Effects of traffic density on mental workload showed individual differences between participants. The model shows that stress and workload are dependent on historic values.
[53]	HR, HRV	On-road	Likon Prince180D heart rate tester (North-vision Tech. Inc.)	Optical vs non-optical tunnels	SDNN and RMSSD comparison	Increased mean HR when driving in tunnels compared to resting state, no significant difference between optical and non-optical tunnels. Non-optical tunnel SDNN was lesser than optical tunnel SDNN. RMDDS decreased in non-optical tunnels compared to optical tunnels.
[30]	HR, HRV, RR	Simulator	Biopac MP150 BioNomadix (BIOPAC Hardware)	Low cognitive vs high cognitive mental arithmetic secondary task	Two-way ANOVA	HR increased for high cognitive tasks. RMSSD decreased, showing opposing effects. pNN20 decreased. No significant results for SDNN. RR increased for the driving condition, with no change in a high cognitive task.
[70]	HR, HRV, RR	Simulator	Biopac MP36 (BIOPAC Hardware)	Verbal cognitive secondary tasks	ANOVA	ECG + RR signals presented the best accuracy for the classifier (92–94%).
[6]	HR, HRV	Simulator	PolymateV AP5148 (Miyuki Giken, AnalyzeDirect Inc.)	Traffic density, pedestrians	ML: Long Short-Term Memory classification	LSTM-based 5-class classification was performed to achieve 96.5% accuracy in relation with subjective measures.
[71]	HR, HRV, RR	On-road	3-lead ECG, respiratory belt (Manufacturer not specified)	N-back tasks	ANOVA, Paired t-test, ML: CNN-LSTM, CNN, Conv-LSTM, XGBoost	CNN and LSTM had the best classification results (97.8%). Four classes: relaxed, normal, high, very high.
[72]	HR, HRV	Simulator	Microvit MT 101 (Schiller)	Lane changing task, secondary N-back task	ANOVA, Post-hoc pairwise comparison	RMSSD, SDSD, pNN20, pNN50, and HF remained constant from rest to level 1 and increased from level 1 to 2. Only three parameters showed difference between levels 2 and 3: LF, HF, pNN50.
[73]	HR, HRV	Simulator	Biopac MP36 (BIOPAC Hardware)	Overtaking events, pedestrians	ANOVA, ML: Random Forest, C-support vector classification, multilayer perceptron	Considering short-term windows, HR and HRV increased in overtaking events and in the presence of pedestrians.