Table 5.
Summary and statistical comparison to other methods to classification alcoholism and healthy patients.
| Author | Data | Channels | Subjects | Features/Method | Classification of features | Performance metrics | |||
|---|---|---|---|---|---|---|---|---|---|
| Acc | Precision | F1 Score | Recall | ||||||
| Proposed method | UCI KDD | 11 | 20 | CNN as feature extractor | MobileNet + SVM RBF | 95.33 ± 1.47 | 95.68 ± 1.31 | 95.25 ± 1.52 | 95.00 ± 1.63 |
| Acharya et al. (2012) | Bern Barcelona database |
2 | 3000 | Entropy, 4 HOS features, Largest Lyapunov Entropy |
SVM (linear, polynomial, and RBF kernels) |
91.7 | 93.9 | – | 90 |
| Rachman et al. (2016) | UCI KDD | 64 | 77 | Daubechies wavelet family | Maximum, minimum, average and standard |
85 | – | – | 100 |
| Mumtaz et al. (2016) | University Malaya Medical Center |
19 | 45 | Power Spectral Density (PSD) | Logistic Regression | 89.5 | 88.5 | 91 | 90 |
| Ehlers et al. (1998) | University of California |
1 | 32 | CD | Discriminant analysis | 88 | – | – | – |
| Kannathal et al. (2005) | UCI KDD | 60 | 30 | CD, LLE, entropy, H | Filter by unique ranges | 90 | – | – | – |
| Faust et al. (2013) | UCI KDD | 61 | 60 | HOS cumulants | FSC | 92.4 | 91.1 | – | 94.9 |
| Patidar et al. (2017) | UCI KDD | 64 | 122 | Tunable Q-wavelet transform | Correntropy, Low-frequency(LF)-rhythms based statistical features |
97.02 | – | – | 96.53 |