Intrusion Detection System for Internet of Things Based on Temporal Convolution Neural Network and Efficient Feature Engineering

Table 1

Deep learning-based IDS for IoT.

Ref	DL technique	Overfitting	Unbalanced dataset	Feature engineering	Model optimization	Testing on IoT dataset
[26]	CNN-RNN	Yes	No	FS	No	No: RedIRIS
[28]	GRU	No	No	FS:RF	Yes	No: KDDCup’99
[31]	BLSTM RNN	No	No	FE	No	No: UNSW-NB15
[24]	DNN	Yes	No	FE	Yes	No: NSL-KDD
[29]	MLP, 1d-CNN, LSTM, CNN+LSTM	Yes	Yes	No	No	No: CICIDS2017
[32]	DNN	No	No	No	Yes	Yes: simulation
[27]	SDPN	Yes	No	FS:SMO	Yes	No: NSL-KDD
[25]	RNN-BPTT	No	No	FN	No	Yes: Bot-IoT
[30]	CNN+LSTM	Yes	No	FS: NSGA	No	No: CISIDS2017
[49]	RNN, LSTM	No	No	FS:CC	No	Yes: Bot-IoT
[3]	DeepDCA	No	Yes	FS:IG	Yes	Yes: BoT-IoT
[51]	ANN	No	SMOTE	FN	No	Yes: BoT-IoT
[52]	FNN	Yes	Yes	FE	Yes	Yes: BoT-IoT
[53]	DAE-DFFFN	No	No	FE, FN	Yes	No: NSL-KDD, UNSW-NB15
Our work	TCNN	Yes	SMOTE-NC	FSR	Yes	Yes: BoT-IoT
				FT: LT, SS, FE

FS: feature selection; RF: random forest; FE: feature encoding; FN: feature normalization; FSR: feature space reduction; IG: information gain; CC: correlation coefficient; FT: feature transformation; LT: log transformation; SS: standard scaler; SMO: Spider Monkey Optimization; NSGA: nondominated sorting genetic algorithm.