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

Table 7

Performance of machine learning models.

Detection model	Oversampling	Phase	Log loss	Accuracy	Precision	Recall	F1-score	Training time (s)
LR	None	Training	0.055841	97.0861%	59.8890%	81.2382%	52.1265%	511
		Testing	0.057109	97.0598%	59.9419%	82.6940%	52.1741%
	SMOTE-NC	Training	0.075336	99.2955%	75.2781%	99.2496%	79.6344%	709
		Testing	0.077694	99.2858%	74.5496%	98.6987%	78.9640%
RF	None	Training	0.200992	97.4837%	80.4852%	98.8858%	86.8911%	191
		Testing	0.20116	97.4586%	77.8298%	98.8643%	84.4592%
	SMOTE-NC	Training	0.195178	96.6396%	79.8083%	98.5854%	86.3145%	197
		Testing	0.195124	96.6341%	75.8464%	98.5543%	82.6850%