Table 5.
Comparative analysis of the performance.
| Model | Accuracy | F1-Score | Input Data | Scenarios | Number of Participants |
|---|---|---|---|---|---|
| Transformer [26] | 71.60% | 74.20% | Raw ECG | Participants write reports for each of the two provided topics and make presentation for one of the provided topics (SWELL dataset) | 25 |
| Random Forest [27] | 78.80% | 88.80% | Extracted features of GSR, heart rate | Students perform multiple tasks, including sing-a-song, emails, color-word test, game, arithmetic question, social conversation, eating, homework, put hands in ice bucket | 9 |
| AdaBoost DT (3-class classification) [17] | 80.34% | 72.51% | Extracted features of PPG, EDA, SKT | Participants read magazines, take TSST, and watch amusing videos (WESAD dataset) | 17 |
| DeepER Net [28] | 83.90% | 81.00% | Extracted features of ECG and RSP | University students solve math tasks or color-word test | 18 |
| Artificial Neural Network (ANN) [29] | 84.32% | 78.71% | Extracted features of ACC, PPG, EDA, TEMP, RESP, EMG, and ECG | Participants read magazines, take TSST, and watch amusing videos (WESAD dataset) | 17 |
| Deep ECGNet [30] | 87.39% | 73.96% | Extracted features of ECG | Students take multiple tasks, including arithmetic problems, color-word test, interview | 30 |
| CNN-LSTM Network [31] | 92.80% | 94.56% | Raw ECG, vehicle dynamic data, environmental parameters | Participants drive on a simulator with different scenarios, including urban, highway, city | 17 |
| Multi-layer Perceptron [24] | 93.64% | 92.44% | Raw PPG, EDA, SKT | Participants read magazines, take TSST, and watch amusing videos (WESAD dataset) | 17 |
| SVM-RBF [25] | 96.25% | 96.00% | Extracted features of PPG, GSR, EEG | Participants prepare a talk and speak in front of real audience | 40 |
| Deep 1D-CNN [12] | 97.48% | 96.82% | ECG, EDA, EMG, RESP, TEMP, TEMP, ACC | Participants watched a series of videos | 15 |
| Proposed model (general) | 93.42% | 88.11% | ECG + EEG | Students solve Sudoku puzzles under different distractions, including noisy environment, another individual monitoring, comforting conditions | 30 |
| Proposed model (scenario 1) | 95.13% | 93.72% | |||
| Proposed model (scenario 2) | 97.76% | 96.67% | |||
| Proposed model (scenario 3) | 98.78% | 95.39% |