Scholarlink

Intrusion detection systems (IDS) that are based on anomalies are crucial for identifying anomalous and dangerous activity in computer networks. In order to identify intrusions and categorize various attack types, we provide ML-machine learning models. The UNSW-NB15 dataset, which includes nine different assault types and 49 attributes, is used in the study. With an reliability of 99%, the Decision Tree outperformed Random Forest (97.97%), AdaBoost (96.80%), and XGBoost (97.086%) among machine learning models. Additionally, the K-Nearest Neighbor (KNN) method was tried for a range of K values. It performed best at K = 7, obtaining an accuracy of 95.58.In terms of attack-wise detection, Decision Trees found reconnaissance (79%), Random Forest identified fuzzers (90%) and generic assaults (99%), and XGBoost identified exploits (95%). There was no requirement for feature selection because every dataset feature made a significant contribution to detection.

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

Next-Generation Network Security: ML Based Attack Detection

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sneha Padhiar,
  • Mehul Barot

摘要

Intrusion detection systems (IDS) that are based on anomalies are crucial for identifying anomalous and dangerous activity in computer networks. In order to identify intrusions and categorize various attack types, we provide ML-machine learning models. The UNSW-NB15 dataset, which includes nine different assault types and 49 attributes, is used in the study. With an reliability of 99%, the Decision Tree outperformed Random Forest (97.97%), AdaBoost (96.80%), and XGBoost (97.086%) among machine learning models. Additionally, the K-Nearest Neighbor (KNN) method was tried for a range of K values. It performed best at K = 7, obtaining an accuracy of 95.58.In terms of attack-wise detection, Decision Trees found reconnaissance (79%), Random Forest identified fuzzers (90%) and generic assaults (99%), and XGBoost identified exploits (95%). There was no requirement for feature selection because every dataset feature made a significant contribution to detection.