Scholarlink

In this study, we evaluate and compare two feature optimization methods for enhancing Intrusion Detection Systems (IDS): Genetic Algorithm (GA)-based and Decision Tree (DT)-based approaches. Both methods aim to select optimal feature subsets to improve classification performance and reduce computational cost. The GA-based approach integrates a novel fitness function with Least Squares Support Vector Machine (LSSVM) to simultaneously maximize the True Positive Rate (TPR), minimize the False Positive Rate (FPR), and optimize features. The DT-based method leverages a hybrid filter-wrapper strategy using C5.0 decision trees and LSSVM, employing gain ratio and pruning for initial selection, followed by predictor importance ranking for refined optimization. Experimental evaluations using the KDD CUP 99 and UNSW-NB15 datasets demonstrate that the DT-based feature selection consistently outperforms the GA-based method across most attack categories in terms of accuracy, TPR, FPR, and Area Under the Receiver Operating Characteristic (ROC) Curve (AUC). The results show that DT has better performance than GA with respect to feature optimization.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Evaluation of Genetic Algorithm and Decision Tree Optimizations for Anomaly Detection IDS

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ali Gharaee,
  • Hossein Gharaee Garakani

摘要

In this study, we evaluate and compare two feature optimization methods for enhancing Intrusion Detection Systems (IDS): Genetic Algorithm (GA)-based and Decision Tree (DT)-based approaches. Both methods aim to select optimal feature subsets to improve classification performance and reduce computational cost. The GA-based approach integrates a novel fitness function with Least Squares Support Vector Machine (LSSVM) to simultaneously maximize the True Positive Rate (TPR), minimize the False Positive Rate (FPR), and optimize features. The DT-based method leverages a hybrid filter-wrapper strategy using C5.0 decision trees and LSSVM, employing gain ratio and pruning for initial selection, followed by predictor importance ranking for refined optimization. Experimental evaluations using the KDD CUP 99 and UNSW-NB15 datasets demonstrate that the DT-based feature selection consistently outperforms the GA-based method across most attack categories in terms of accuracy, TPR, FPR, and Area Under the Receiver Operating Characteristic (ROC) Curve (AUC). The results show that DT has better performance than GA with respect to feature optimization.