Sign in Subscribe
Machine learning

Enhanced Network Traffic Classification Using Bayesian-Optimized Logistic Regression and Random Forest Algorithm

Strategic Insights

— 1 min read

Research at Vellore Institute of Technology has highlighted the need for effective real-time network security solutions in the face of increasing cyber threats. By employing Bayesian optimization to fine-tune the hyperparameters of machine learning models, researchers have developed more accurate and efficient methods for detecting TOR traffic. These models, which combine logistic regression and random forest algorithms, demonstrate superior performance in balancing accuracy, computational efficiency, and detection speed.

Key Takeaways:

  • The research explores the use of Bayesian optimization to improve the performance of machine learning models in detecting TOR traffic.
  • The proposed models, which combine logistic regression and random forest algorithms, achieve higher accuracy and fewer false positives compared to other machine learning algorithms.
  • The approach's flexibility ensures consistent optimal performance as traffic patterns change.
  • The study evaluates four key machine learning algorithms: Random forest, logistic regression, support vector machine (SVM), and K-nearest neighbors (K-NN).
  • Bayesian optimization is employed to systematically fine-tune the model's hyperparameters, improving accuracy and efficiency by concentrating on promising areas of the hyperparameter space and avoiding unnecessary evaluations.
  • The findings indicate that the proposed models are ideal for real-time network security applications.
  • The research was led by Manisankar Sannigrahi at Vellore Institute of Technology.

Statistics:

  • The proposed models achieve an accuracy of 95% in detecting TOR traffic using the UNB-CIC TOR-NonTOR datasets.
  • The models demonstrate a 30% reduction in false positives compared to other machine learning algorithms.
  • The study evaluates the performance of the proposed models on two datasets: UNB-CIC TOR-NonTOR and CIC-Darknet2020.
  • The research concludes that the approach's flexibility ensures consistent optimal performance as traffic patterns change.

Sources:

  • Enhanced Network Traffic Classification Using Bayesian-Optimized Logistic Regression and Random Forest Algorithm. Journal of Engineering. 2025
  • Vellore Institute of Technology
  • Journal of Engineering - https://www.hindawi.com/journals/je/
  • Wiley

Read more