Table 5.
Results of multilayer perceptron with varied hyperparameter settings.
Hidden layers (n) | Hidden layer size | BoWa | Activation | Epochs (n) | Recall | Precision | F1-score | Accuracy | AUCb |
2 | 50, 50 | No | Rectifier | 10 | 90.28 | 96.25 | 93.17 | 90.2 | 0.967 |
3 | 100, 50, 25 | No | Rectifier | 10 | 93.47 | 96.71 | 95.06 | 92.8 | 0.969 |
3 | 100, 50, 25 | No | Maxout | 10 | 96.82 | 96.01 | 96.41 | 94.67 | 0.976 |
3 | 100, 50, 25 | No | Maxout with Dropout | 10 | 98.16 | 93.61 | 95.83 | 93.67 | 0.963 |
3 | 100, 50, 25 | No | Tanh | 10 | 90.62 | 97.65 | 94.00 | 91.44 | 0.978 |
3 | 100, 50, 25 | Yes | Rectifier | 10 | 93.47 | 98.24 | 95.80 | 93.92 | 0.999 |
3 | 100, 50, 25 | Yes | Maxout | 10 | 94.47 | 97.41 | 95.92 | 94.04 | 0.977 |
3 | 50, 50, 50 | No | Rectifier | 10 | 87.77 | 96.86 | 92.09 | 88.83 | 0.958 |
3 | 200, 100, 50 | No | Rectifier | 10 | 92.96 | 96.86 | 94.87 | 92.56 | 0.975 |
4 | 200, 100, 50, 25 | No | Rectifier | 10 | 93.63 | 96.05 | 94.82 | 92.43 | 0.973 |
aBoW: bag of words.
bAUC: area under the receiver operating characteristic curve.