•  
  •  
 

Document Type

Original Study

Subject Areas

Information science

Keywords

Intrusion Detection System (IDS); Principal Component Analysis (PCA); Genetic Algorithm (GA); Deep Neural Networks (DNNs)

Abstract

With cyber-attacks on the rise, the demand for precise and efficient Intrusion Detection Systems (IDS) is more critical than ever. Traditional machine learning methods struggle with high false alarm and poor detection rates. IDS are crucial for creating secure digital environments by classifying and responding to network intrusions. However, the large volume of inputs for classification often leads to reduced accuracy and misclassification errors. To address this problem, Principal Component Analysis (PCA) is used to optimize the number of classifier inputs, enhancing simplicity without compromising accuracy. A significant challenge in PCA for feature selection lies in determining the best cutoff of principal components to identify the most relevant features. This paper introduces a hybrid approach using the Genetic Algorithm (GA) with PCA and Deep Neural Networks (DNNs) to refine feature selection and find the optimum cutoff that minimizes the number of features while maximizing accuracy. In addition, for comparative evaluation, a DNN with L1-based feature selection (L1-FS-DNN) was employed as a benchmark baseline to assess the effectiveness of the proposed framework. Experiments conducted on the CICIDS2017 dataset show that the proposed PCA-GA-DNN approach reduced the feature set by approximately 73.8% and achieved a classification accuracy of 98.86%, outperforming the comparative baseline in terms of accuracy, feature reduction capability, and training stability. These findings demonstrate the effectiveness of the proposed method in enhancing IDS performance.

Download

Share

COinS