Class Imbalance in Network Traffic Classification An Adaptive Weight Ensemble-of-Ensemble Learning Method

Network Traffic Classification (NTC) plays a vital role in cybersecurity by enabling the identification and analysis of network data patterns. Traditional methods relying on port or payload inspection face challenges due to encryption and dynamic port usage. Machine learning techniques have emerged as effective alternatives, but they often struggle with class imbalance, where benign traffic dominates over attack data. This imbalance reduces the detection accuracy for critical minority classes such as attacks. This project addresses the class imbalance issue by proposing an adaptive ensemble learning method that combines multiple classifiers into a stacking framework. The system integrates Decision Trees, Random Forests, and XGBoost as base learners, with a meta-learner to refine final predictions. Additionally, explainable artificial intelligence techniques using SHAP are employed to enhance model transparency and trustworthiness. The proposed approach also incorporates feature engineering and data preprocessing steps to optimize performance. Evaluation metrics such as Accuracy, Precision, Recall, and F1-Score are used to assess the system’s effectiveness. Overall, this adaptive weighted ensemble improves detection of minority traffic classes, offers interpretability, and shows strong potential for scalable and robust network traffic classification in cybersecurity applications.

Keywords: Network Traffic Classification, Class Imbalance, Ensemble Learning, Stacking Classifier, Decision Tree, Random Forest, XGBoost, SHAP, Feature Engineering, Cybersecurity