FS Based on HEVO: A Novel Energy Valley-Based Optimization Framework for Efficient Feature Selection in AI-Driven Cybersecurity
Year: 2024
Type: conference-paper
Abstract: Cybersecurity presents a significant challenge for Intrusion Detection Systems (IDSs) due to the large size and high dimensionality of datasets, compounded by skewed data distributions and poor classification model performance. This article presents a novel Energy Valley Optimizer (EVO)-based feature selection (FS) and dimensionality reduction approach that improves IDS effectiveness. The CSE-CIC-IDS2018 dataset is huge and highly dimensional, so EVO was chosen to reduce the feature set to make IDS models more efficient and less computationally costly. The proposed EVO technique is employed for feature selection to illustrate its efficacy in enhancing IDS input data, model efficiency, and classification outcomes. The suggested method uses Support Vector Machine (SVM), Random Forest (RF), Decision Tree (D_Tree), and K-Nearest Neighbors (KNN) machine learning (ML) models. Downsampling improves model accuracy in distinguishing benign and malicious traffic. EVO decreased features from 80 to 43 in our testing, boosting classification accuracy, precision, and recall across all models. These results show that EVO's specific feature selection technique reduces data complexity and boosts IDS detection rates. Findings imply that EVO improves feature selection, IDS performance to protect cybersecurity domains
Author: Alhusseini
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7167802449