Research Article
Empirical Evaluation of Noise Influence on Supervised Machine Learning Algorithms Using Intrusion Detection Datasets
Table 15
The results of running ML algorithms on a filtered and then injected intrusion dataset.
| Noise (%) | ML algorithm | NSL-KDD | UNSW-NB15 | Accuracy | Precision | Recall | Accuracy | Precision | Recall |
| Baseline | DT (J48) | 81.5339 | 0.858 | 0.815 | 100 | 1.000 | 1.000 | RF | 80.4516 | 0.852 | 0.805 | 98.4903 | 0.985 | 0.985 | SVM | 75.3948 | 0.802 | 0.754 | 100 | 1.000 | 1.000 | ANN | 77.7147 | 0.817 | 0.777 | 100 | 1.000 | 1.000 | NB | 76.1178 | 0.809 | 0.761 | 87.4435 | 0.884 | 0.874 |
| Filtered then 5% noise injected | DT (J48) | 80.6742 | 0.844 | 0.807 | 100 | 1.000 | 1.000 | RF | 80.4869 | 0.746 | 0.805 | 96.3296 | 0.967 | 0.963 | SVM | 80.4869 | 0.746 | 0.805 | 100 | 1.000 | 1.000 | ANN | 80.5243 | 0.760 | 0.805 | 98.5001 | 0.985 | 0.985 | NB | 80.4869 | 0.746 | 0.805 | 74.285 | 0.795 | 0.743 |
| Filtered then 10% noise injected | DT (J48) | 80.6742 | 0.844 | 0.807 | 100 | 1.000 | 1.000 | RF | 80.2996 | 0.721 | 0.803 | 96.336 | 0.967 | 0.963 | SVM | 80.4869 | 0.746 | 0.805 | 100 | 1.000 | 1.000 | ANN | 80.6367 | 0.796 | 0.806 | 99.892 | 0.999 | 0.999 | NB | 80.4869 | 0.746 | 0.805 | 74.2055 | 0.795 | 0.742 |
| Filtered then 20% noise injected | DT (J48) | 80.5618 | 0.843 | 0.806 | 71.4472 | 0.815 | 0.714 | RF | 81.2734 | 0.777 | 0.813 | 92.656 | 0.938 | 0.927 | SVM | 80.4869 | 0.746 | 0.805 | 99.9841 | 1.000 | 1.000 | ANN | 80.6367 | 0.796 | 0.806 | 99.9682 | 1.000 | 1.000 | NB | 80.4869 | 0.746 | 0.805 | 74.0784 | 0.794 | 0.741 |
| Filtered then 30% noise injected | DT (J48) | 78.5768 | 0.662 | 0.786 | 70.1824 | 0.797 | 0.702 | RF | 80.9738 | 0.804 | 0.810 | 87.6001 | 0.897 | 0.876 | SVM | 80.4869 | 0.746 | 0.805 | 99.9174 | 0.999 | 0.999 | ANN | 80.7116 | 0.766 | 0.807 | 99.9078 | 0.999 | 0.999 | NB | 80.4869 | 0.746 | 0.805 | 74.0721 | 0.794 | 0.741 |
|
|