BalancedBoost: A hybrid approach for real-time network traffic classification

Real-time network traffic classification suffers from class imbalance and concept drift problem. Moreover, many features for offline analysis becomes unavailable due to the time requirement of giving results as quickly as possible. Several techniques, such as boosting and data sampling, have been used to solve the class imbalance problem. In this paper, we proposed a hybrid approach, called BalancedBoost, to address the class imbalance problem. BalancedBoost is based on a combination of data sampling and boosting algorithm. In order to reduce the impact of traffic dynamics and filter the redundant features, a feature selection algorithm based on symmetrical uncertainty is also presented. We conducted experiments on UNIBS datasets and compared our method to other similar proposed approaches including SMOTEBoost, RUSBoost and OverBagging. Experiment results obtained by our method are promising in terms of F-measure. BalancedBoost is able to improve the classification performance of the minority class. It is simple and efficient.