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

Table 14

Injecting different levels of noise in the intrusion datasets.

Noise (%)	ML algorithm	NSL-KDD			UNSW-NB15
		Accuracy	Precision	Recall	Accuracy	Precision	Recall
Baseline 0% noise	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
5% noise injected	DT (J48)	80.7887	0.851	0.808	100	1.000	1.000
	RF	79.3204	0.844	0.793	94.8538	0.954	0.949
	SVM	76.2731	0.828	0.763	100	1.000	1.000
	ANN	79.2273	0.826	0.792	99.9988	1.000	1.000
	NB	74.9867	0.805	0.750	77.5992	0.805	0.776
10% noise injected	DT (J48)	80.7976	0.846	0.808	100	1.000	1.000
	RF	77.2179	0.814	0.772	93.9829	0.947	0.940
	SVM	76.2376	0.828	0.762	100	1.000	1.000
	ANN	76.4061	0.809	0.764	100	1.000	1.000
	NB	73.9354	0.800	0.739	77.5883	0.801	0.776
20% noise injected	DT (J48)	79.8084	0.826	0.798	100	1.000	1.000
	RF	72.3208	0.777	0.723	88.8148	0.908	0.888
	SVM	78.7615	0.848	0.788	100	1.000	1.000
	ANN	77.4308	0.813	0.774	99.9988	1.000	1.000
	NB	71.6776	0.791	0.717	77.2713	0.800	0.773
30% noise injected	DT (J48)	80.1011	0.828	0.801	66.9412	0.778	0.669
	RF	70.6574	0.750	0.707	84.4799	0.871	0.845
	SVM	78.7216	0.847	0.787	100	1.000	1.000
	ANN	75.2928	0.802	0.753	99.9818	1.000	1.000
	NB	74.4677	0.801	0.745	77.7511	0.801	0.778