Table 7.
Comparisons results based on 4-class based DFD systems by using hybrid deep learning based classifier (CNN and RNN–LSTM) along with traditional machine learning (SVM, ANN) and 12 different subjects and recorded time is 30 min.
| Classifiers | AL | VL | MD | ED |
|---|---|---|---|---|
| ANN | SE: 65.6, SP: 67.5, PR: 0.64, ACC: 67 | SE: 66.2, SP: 67, PR: 0.65, ACC: 68 | SE: 67, SP: 68.3, PR: 0.65, ACC: 68 | SE: 75.3, SP: 76.4, PR: 0.75, ACC: 76.5 |
| SVM | SE: 81.3, SP: 82.2, PR: 0.80, ACC: 81 | SE: 80.0, SP: 81.5, PR: 0.81, ACC: 80 | SE: 71.2, SP: 72.3, PR: 0.70, ACC: 71 | SE: 77.1, SP: 78.1, PR: 0.78, ACC: 79.5 |
| CNN+ANN | SE: 82.6, SP: 83.4, PR: 0.82, ACC: 82 | SE: 80.4, SP: 81.3, PR: 0.82, ACC: 81 | SE: 72.4, SP: 73.5, PR: 0.73, ACC: 72 | SE: 78.4, SP: 79.1, PR: 0.78, ACC: 79.0 |
| CNN+SVM | SE: 81.3, SP: 82.2, PR: 0.80, ACC: 81 | SE: 84.0, SP: 85.5, PR: 0.83, ACC: 84 | SE: 78.2, SP: 79.3, PR: 0.78, ACC: 77 | SE: 80.1, SP: 81.1, PR: 0.81, ACC: 81.5 |
| CNN with soft-max classification | SE: 82, SP: 83, PR: 0.83, ACC: 83 |
SE: 84, SP: 85, PR: 0.84, ACC: 84 | SE: 81.2, SP: 82.3, PR: 0.84, ACC: 84 | SE: 84.5, SP: 0.85, PR: 0.84, ACC: 85 |
| CNN+RNN-LSTM | SE: 86.3, SP: 87.6, PR: 0.85, ACC: 86 | SE: 88.3, SP: 89, PR: 0.89, ACC: 89 | SE: 90.0, SP: 91.2, PR: 0.90, ACC: 90 | SE: 92, SP: 93, PR: 0.91, ACC: 92 |
AL: alert, VL: very alert; MD: moderately drowsy, ED: extremely drowsy, SE: sensitivity, specificity: SP, PR: precision, ACC: detection accuracy.