A Novel Framework Design of Network Intrusion Detection Based on Machine Learning Techniques

Table 9

Comparison of the proposed framework in the multiclass classification with related work using CICIDS2017 dataset evaluation and some of the traditional ML approaches mentioned in their papers.

	Reference + method	Accuracy	Precision	Recall	F1-score	Remark
[8]	MDAE + LSTM	0.9860	0.9860	0.9960	0.9860	8 classification
	Naive Bayes	0.2500	0.7670	0.2500	0.1880
	SVM	0.7990	0.7570	0.7990	0.7230
	DNN	0.9480	0.9650	0.9480	0.9530
	MDAE	0.9040	0.9920	0.9000	0.9110
	LSTM	0.9700	0.9680	0.9860	0.9730
[17]	DT + rule-based	0.9967		0.9448		15 classification
	RF	0.9559		0.9305
	REP tree	0.9340		0.9164
	Multilayer Perceptron	0.8524		0.7783
	Naive Bayes	0.7453		0.8251
	Jrip	0.9447		0.9340
	J48	0.9348		0.9199
[13]	DBN-SVM		0.9774	0.9768	0.9768	Use only Tuesday’s dataset with 5 classification
[9]	CNN + LSTM	0.9867		0.9721	0.9332	7 classification
	CNN	0.9844		0.9646	0.9311
	LSTM	0.9683		0.9421	0.9097
[23]	PCA + RF	0.9880	0.9890	0.9880	0.9880	15 classification
	AE + RF	0.9950			0.9950
	Our approach	0.9990	0.9990	0.9989	0.9989	15 classification

Note. The best values are in bold, and missing value means not provided in reference.