EFS-DNN: An Ensemble Feature Selection-Based Deep Learning Approach to Network Intrusion Detection System

Algorithm 1

EFS-DNN training procedure.

	Input: The original dataset
	Input: The subset sample rate
	Input: The feature selection threshold
	Input: The number of LightGBMs
	Input: The number of epochs
	Output: The classification results
	Step 1: Pre-process the original dataset
	Perform Min-Max normalization: ;
	Step 2: Calculate feature importance
	Initialize the feature importance: ;
	Fortodo:
	Construct subset according to sample rate: ;
	Train LightGBM based on the subset: ;
	Accumulate feature importance: ;
	End
	Step 3: Select features by threshold
	Sort feature importance in descending: ;
	Select features according to threshold: ;
	Step 4: Construct DNN classifier
	Separate features into categorical and numerical features: ;
	Map categorical features to dense vector: ;
	Concatenate embedding and numerical features: ;
	Define deep neural network: ;
	Step 5: Train deep neural network
	Fortodo:
	Predict : ;
	Calculate cross-entropy loss: ;
	Optimize model by Adam optimizer;
	End
	Return