Evolutionary Data Preprocessing to Alleviate Class Imbalance

Table 10

Experimental results of the training dataset preprocessed to the best tuple of SMOTE ratios, validation, and test dataset.

			D-S-1G	R-R-2	R-R-2G	S-R-1G
Class	Normal		0.986	0.985	0.986	0.985
	Exploits		0.718	0.366	0.364	0.41
	Reconnaissance		0.704	0.665	0.658	0.658
	DoS		0.321	0.011	0.009	0.705
	Generic		0.897	0.972	0.972	0.972
	Shellcode		0.742	0.372	0.385	0.416
	Fuzzers		0.866	0.357	0.357	0.506
	Worms		0.534	0.931	0.914	0.948
	Backdoor		0.357	0.108	0.061	0.309
	Analysis		0.206	0.975	0.979	0.204
Result	W. Avg.	Recall	0.965	0.693	0.959	0.760
		Precision	0.979	0.858	0.980	0.866
		F-measure	0.971	0.726	0.966	0.785
			1.232	1.350	1.338	1.223
			0.196	0.676	0.662	0.508
			0.266	0.378	0.384	0.287
			0.196	0.454	0.463	0.304

Bold values mean the rare classes.