Standardizing Preprocessing for Machine Learning-Based Network Intrusion Detection Systems

Strategic Insights -

Researchers at the University of Queensland have identified a significant issue in the field of machine learning-based network intrusion detection systems, where inconsistencies in data preprocessing methods are hindering fair comparison and limiting performance. Despite the extensive efforts in developing these systems, the choice of pre-processing of training data varies significantly across the literature, often without in-depth discussion and justification. To address this problem, the researchers propose a standardized pre-processing approach for Network Intrusion Detection Systems benchmark datasets, which consistently performs well across different machine learning models.

Key Takeaways:

  • The research highlights the need for standardization in data preprocessing methods for machine learning-based network intrusion detection systems.
  • The existing literature shows significant variations in pre-processing methods, often without justification, resulting in inconsistent performance.
  • The proposed standardized pre-processing approach consistently performs well across different machine learning models and datasets.
  • The researchers performed a detailed experimental evaluation on 72 combinations of pre-processing methods for both numerical and categorical fields in network flow data.
  • The proposed method can achieve up to an 11% increase in detection accuracy over commonly applied preprocessing techniques for shallow neural networks.
  • The study makes the code publicly available to facilitate future research in the field.
  • The research has been peer-reviewed and published in the Engineering Applications of Artificial Intelligence journal.

Statistics:

  • 72 combinations of pre-processing methods were evaluated in the study.
  • The proposed standardized pre-processing approach can achieve up to an 11% increase in detection accuracy.
  • The research focused on shallow neural networks with fewer than 3 hidden layers.
  • The study was performed on four prevalent Network Intrusion Detection Systems benchmark datasets.

Sources:

  • NewsRx. Studies from University of Queensland Yield New Information about Machine Learning (An Empirical Evaluation of Preprocessing Methods for Machine Learning Based Network Intrusion Detection Systems). Information Technology Newsweekly. October 21, 2025
  • Engineering Applications of Artificial Intelligence, 2025;158
  • An Empirical Evaluation of Preprocessing Methods for Machine Learning Based Network Intrusion Detection Systems.