Empirical Evaluation of Noise Influence on Supervised Machine Learning Algorithms Using Intrusion Detection Datasets

Table 12

Constructing the baseline experiments with different testing modes.

Test mode	ML algorithm	NSL-KDD			UNSW-NB15
		Accuracy	Precision	Recall	Accuracy	Precession	Recall
Supplied test set	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
	Naïve Bayes	76.1178	0.809	0.761	87.4435	0.884	0.874
Percentage split	DT (J48)	99.7245	0.997	0.997	100	1.000	1.000
	RF	99.9019	0.999	0.999	100	1.000	1.000
	SVM	97.4574	0.975	0.975	100	1.000	1.000
	ANN	98.6085	0.986	0.986	100	1.000	1.000
	Naïve Bayes	90.5933	0.907	0.906	92.1397	0.930	0.921
Cross-validation	DT (J48)	99.7817	0.998	0.998	100	1.000	1.000
	RF	99.9174	0.999	0.999	100	1.000	1.000
	SVM	97.405	0.974	0.974	100	1.000	1.000
	ANN	98.5378	0.985	0.985	100	1.000	1.000
	Naïve Bayes	90.3829	0.905	0.904	92.4809	0.932	0.925