Table 7.

Comparison between the proposed algorithm and existing works for driver stress prediction.

Work	Nature of Dataset	Dataset	Features	Methodology	Time of In-Advance Prediction (s)	Cross-Validation	Results
[19]	Simulated	27 participants; 20,160 samples	Contextual data; vehicle data; ECG	CNN; LSTM	5	No	Accuracy = 92.8%
[20]	Simulated	27 participants; 20,160 samples	Environmental data; vehicle dynamics; eye data	CNN; LSTM; self-attention mechanism	5	10-fold	Accuracy = 95.5%
[21]	Simulated	3 participants; 150 normal samples and 150 stressed samples	HRV; speed and intensity of turning of vehicle	DBN	60	10-fold	Specificity = 62.7–83.6%; Sensitivity = 61.7–82.3%
[22]	Simulated	5 participants; unspecified samples	HRV; weather	NB	30	10-fold	Accuracy = 78.3%
[23]	Real-world	1 participant; 64 low stress samples and 75 high stress samples	Accelerometer; EDA; PPG	LR	60	10-fold	Specificity = 86.7%; Sensitivity = 60.9%
Proposed	Real-world	18 participants; 40,000 LSL samples, 38,000 MSL samples, 16,000 HSL samples	ECG	NSGA-III optimized RNN-GRU-LSTM algorithm	1–60	10-fold	Accuracy = 71.2–93.1%

Convolutional neural network (CNN); deep belief network (DBN); electrocardiogram (ECG); electrodermal activity (EDA); heart rate variability (HRV); high stress level (HSL); logistic regression (LR); low stress level (LSL); long short-term memory (LSTM); medium stress level (MSL); Naive Bayes (NB); photoplethysmography (PPG).