A Novel Hybrid Feature Selection with Cascaded LSTM: Enhancing Security in IoT Networks

Algorithm 1

Stage wise feature selection.

Feature selection Stage 1: RFE Algorithm

1. Initialization:

 I. Feature set: Let F denotes the features where F = {f1, f2,…. fn}, n is the total no of features.

 II. Initialize: The selected the features F´ with all the features of F.

 III. Specify: Let k be the number of features to be selected.

 IV. Elimination: Let D represents the list of eliminated features.

2. Feature Selection:

 I. Perform: Until the features in F´ is greater than k perform: 1–4

  1. Train the RF model with the features of F´.

  2. Compute the feature importance score of the features in F´ as IS(f).

  3. Rank the features of F´ Rank(fi) based on the importance score IS(f).

  4. Remove the least important n–k features from F´. Update the eliminated features in D.

3. RFE Selected Features:

 I. Initial Elimination: The features in F´ denote the selected features and D provides the eliminated features.

Feature selection Stage 2: IG Algorithm

4. Initialization:

 I. Feature set: Let F´ denotes the input dataset with F´ = {f1, f2,… fn–k}.

 II. Initialize: The selected features F´´ with all the features of F´.

 III. Specify: Let k´ be the number of features to be selected.

 IV. Elimination: Let D´ represents the features eliminated using IG.

5. Feature Selection:

 I. Compute the entropy of features in F´´.

 II. Compute the information gain IG of each feature in F´´.

 III. Rank the features of F´´ using the IG score.

 IV. Remove the least important n–k–k´ features from F´´. Update the eliminated features in D´.

6. RFE-IG Selected Features:

 I. Final Elimination: The features in F´´ denotes the selected features and D´ provides the eliminated features.