Table 3.
Evaluation metrics table of only male voice samples for classification task of abnormal voice signals in laryngeal cancer (n = 50) from normal healthy subjects (n = 45).
| Algorithms | Accuracy (%) | Sensitivity (%) | Specificity (%) | AUC |
|---|---|---|---|---|
| SVM | 70.5 (67.9–73.0) | 78.0 (75.6–80.3) | 62.2 (58.0–66.3) | 0.708 |
| XGBoost | 70.5 (68.2–72.8) | 62.0 (58.5–65.4) | 80.0 (77.3–82.6) | 0.731 |
| LightGBM | 71.5 (68.2–74.8) | 70.0 (66.6–73.3) | 73.3 (69.9–76.6) | 0.739 |
| ANN | 69.4 (67.6–71.2) | 62.0 (60.4–63.5) | 77.7 (75.3–80.2) | 0.744 |
| 1D-CNN | 85.2 (83.8–86.6) | 78.0 (76.0–79.9) | 93.3 (92.2–94.4) | 0.852 |
| 2D-CNN * (MFCCs) | 73.3 (72.0–74.7) | 69.6 (66.9–72.2) | 77.5 (74.2–80.8) | 0.778 |
| 2D-CNN * (STFT) | 67.1 (65.6–68.6) | 58.6 (55.6–61.5) | 76.6 (75.1–78.2) | 0.707 |
Abbreviations: AUC, area under curve; SVM, support vector machine; XGBoost, extreme gradient boosting; LightGBM, light gradient boosted machine; ANN, artificial neural network; MFCCs, Mel-frequency cepstral coefficients; STFT, short-time Fourier transform. *: with 10 times augmented data.