Multiclass imbalanced and concept drift network traffic classification framework based on online active learning

Weike Liu, Cheng Zhu, Zhaoyun Ding, Hang Zhang, Qingbao Liu

Abstract: Highlights•An online active learning framework is proposed for multiclass imbalanced network traffic with concept drift.•A novel uncertain label request strategy based on the variable least confidence threshold vector is designed.•A sample weight formula is proposed, which tends to assign larger weights to indistinguishable and minority class samples.•An adaptive adjustment mechanism of the label cost budget based on periodic performance evaluation.•The proposed framework is composed of several separate stages that can be easily modified or tuned.