MidSiot: A Multistage Intrusion Detection System for Internet of Things

Table 6

The results of examined classifiers.

Model	Classification	Accuracy (%)	Training time	Prediction time	Adjusted parameters
Linear support vector machine	Binary	98.16	1150.57	204.54	kernel = linear; gamma = auto
	Multiclass	N/A	10 298.3	1453.1
Quadratic support vector machine	Binary	98.25	792.61	311.5	kernel = poly; gamma = auto
	Multiclass	N/A	N/A	N/A
K-Nearest neighbor	Binary	99.79	0.17	2343.72	n_neighbors = 5
	Multiclass	98.61	0.19	2377.81
Linear discriminant analysis	Binary	95.07	21.6	0.27	All default
	Multiclass	80.73	25.71	4.35
Quadratic discriminant analysis	Binary	53.6	18.89	12.44	All default
	Multiclass	56.62	16.62	14.74
Multilayer perceptron	Binary	99.6	2.57	0.65	Input layer and first layer with 50 neurons and activation = relu; Output layer with activation = sigmoid
	Multiclass	92.71	4.94	7.96
Long short-term memory	Binary	96.53	572.44	62.64	input layer and LSTM layer with 50 neurons; Output layer with activation = sigmoid
	Multiclass	N/A	N/A	N/A
Autoencoder classifier	Binary	93.01	11.65	0.62	Encoding layer with 50 neurons and activation = relu; Decoding and output layer with activation = softmax
	Multiclass	87.74	13.35	0.81
Decision tree classifier	Binary	99.94	12.49	0.43	criterion = entropy
	Multiclass	99.69	16.76	0.38