Dimension Reduction Technique Based on Supervised Autoencoder for Intrusion Detection of Industrial Control Systems

Algorithm 1

Training process of dimension reduction module.

	Data: Training features xi with label yi. Hyperparameter α, the number of iteration t.
	Result: Parameters of dimension reduction module and reduced features
	// Step 1: Preprocessing the training dataset
(1)	Normalize data xi with Min-Max normalization by
	// Step 2: Training SupervisedAE with the normalized dataset
(2)	while not converge do
(3)
(4)	Compute the joint loss by
(5)	Train SupervisedAE using the joint loss and update the parameters
(6)	End
	// Step 3: Computing the latent representation z i by encoder function
(7)
	// Step 4: Reducing the dimension of latent representations z by PCA
(8)