Automated Hyperparameter Tuning and Ensemble Machine Learning Approach for Network Traffic Classification

Network traffic analysis and forecasting have become indispensable to dynamically allocate bandwidth, congestion control, security, and planning in today’s complex, heterogeneous, and traffic-intensive networks. This study focuses on leveraging and optimizing machine learning techniques to perform network traffic analysis and classification for decision-making. The study initiated with a thorough exploration of machine learning algorithms. Various algorithms, including Logistic Regression, SVM, KNN, Decision Trees, Random Forests, Extra Trees, Bagging, and AdaBoost, are applied for multi-class classification in network traffic analysis. To improve accuracy, we first employed automated hyperparameter tuning for each of the algorithms. After that we implemented the ensemble method with unanimous voting. In terms of unanimous voting, increasing the number of algorithms with certain criteria correlates with a decrease in false positives and false negatives. The results of our approach clearly improve the accuracy of the network traffic classification over fundamental machine learning models for better management of complex heterogeneous networks.