Table 3.
Predictive performance of classifying electron transport proteins using different neural networks
| CV | Independent | |||||||
|---|---|---|---|---|---|---|---|---|
| Sen | Spe | Acc | MCC | Sen | Spe | Acc | MCC | |
| kNN | 37.7(−) | 98.9(+) | 85.2(−) | 0.53(−) | 32.7(−) | 96.5(+) | 82.1(−) | 0.41(−) |
| RF | 64.8(−) | 97.1(+) | 89.8(−) | 0.69(−) | 56.3(−) | 96.4(+) | 87.3(−) | 0.61(−) |
| SVM | 74(−) | 96.2(+) | 91.2(−) | 0.74(−) | 74(−) | 91.7(−) | 87.7(−) | 0.65(−) |
| CNN | 73.8(−) | 95(−) | 90.3(−) | 0.71(−) | 78.2(+) | 92.5(−) | 89.5(−) | 0.69(−) |
| GRU | 83.7 | 96.3 | 93.5 | 0.81 | 79.8 | 95.9 | 92.3 | 0.77 |
Note: (kNN: k = 10, RF: n_estimators = 500, SVM: c = 32, g = 0.125, CNN: 128 filters, GRU: 32 filters, (+) for significantly better than GRU, (−) for significantly worse than GRU in a two-proportion z-test)