Scholarlink

Cybersecurity has emerged as one of the major concerns in digital transformation. The protection of digital assets and maintenance of data integrity could only be ensured with the appropriate and timely prediction of these attacks. In this paper, we present an investigation of how different feature selection techniques affect Machine Learning (ML) model performance in predicting cyber-attacks. The proposed Intrusion Detection System (IDS) consists of some phases starting from the preprocessing phase, the feature selection stage to the detection stage. Feature selection is one of the vital steps in machine learning, as it reduces dimensionality by the elimination of noisy and redundant data, which enhances model accuracy by identifying only the most relevant features. This work evaluates various methods of feature selections applied, including Analysis of Variance (ANOVA), Chi-square, and Mutual Information (MI) techniques, and their impact on the performance of different machine learning algorithms like Naïve Bayes (NB), K-Nearest Neighbor (KNN), AdaBoost, Random Forest (RF) and Decision Tree (DT). Evaluated the proposed approach on an NSLKDD dataset yielded promising results, with accuracy ranging from 99.9% for the time series dataset. In conclusion, the proposed approach effectively detects and mitigates DoS attacks, offering a reliable solution to protect communication systems against this growing cybersecurity threat.

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

Influence of Feature Selection Techniques on the Cyber-Attacks Prediction Using Machine Learning Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Hussein Salam Ali,
  • Ali Hasan Taresh

摘要

Cybersecurity has emerged as one of the major concerns in digital transformation. The protection of digital assets and maintenance of data integrity could only be ensured with the appropriate and timely prediction of these attacks. In this paper, we present an investigation of how different feature selection techniques affect Machine Learning (ML) model performance in predicting cyber-attacks. The proposed Intrusion Detection System (IDS) consists of some phases starting from the preprocessing phase, the feature selection stage to the detection stage. Feature selection is one of the vital steps in machine learning, as it reduces dimensionality by the elimination of noisy and redundant data, which enhances model accuracy by identifying only the most relevant features. This work evaluates various methods of feature selections applied, including Analysis of Variance (ANOVA), Chi-square, and Mutual Information (MI) techniques, and their impact on the performance of different machine learning algorithms like Naïve Bayes (NB), K-Nearest Neighbor (KNN), AdaBoost, Random Forest (RF) and Decision Tree (DT). Evaluated the proposed approach on an NSLKDD dataset yielded promising results, with accuracy ranging from 99.9% for the time series dataset. In conclusion, the proposed approach effectively detects and mitigates DoS attacks, offering a reliable solution to protect communication systems against this growing cybersecurity threat.