What is it about?
This work highlights the importance of tuning the HPs of ML learning models to build effective network intrusion detection systems. It compares the performance of three different hyper-parameter optimization algorithms (HPO) for tuning the HPs of ML algorithms. The selection decision of which HPO algorithms to use was made such that it includes three different optimization techniques: a model-based approach, a multi-fidelity approach, and a heuristic approach. Since some of the public network intrusion detection (NID) datasets suffer from imbalance issues, we propose an algorithm for data balancing to enhance the performance of ML/DL algorithms where a dynamic approach that considers both under-sampling and over-sampling is applied.
Featured Image
Photo by Joshua Sortino on Unsplash
Why is it important?
In our work, we show how HP optimization can significantly improve the performance of machine learning and deep learning algorithms. Our optimized models achieved a recall of 100% on the BoT-IoT dataset and a recall of 99.96% on the ToN-IoT dataset. With our proposed data balancing algorithm, we have managed to significantly improve the false alarm rate (FAR) of the deep neural network (DNN) model from 30% to 0% while maintaining a 0% false negative rate (FNR).
Perspectives
The findings in this work can help ML practitioners and researchers define search spaces for their HP optimization problems with the confidence that optimization algorithms can converge given decent resource budgets. Furthermore, the data balancing algorithm can serve as a base for building balancing strategies that adopt other data sampling approaches.
Omar Elghalhoud
University of Waterloo
Read the Original
This page is a summary of: Data Balancing and Hyper-parameter Optimization for Machine Learning Algorithms for Secure IoT Networks, October 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3551661.3561364.
You can read the full text:
Contributors
The following have contributed to this page
