Classifying Imbalanced Data Streams via Dynamic Feature Group Weighting with Importance Sampling.

Area of Science:

• Data mining
• Machine learning
• Big data analytics

Background:

• Data stream classification and imbalanced data learning are critical research areas.
• Few methods effectively address the intersection of these fields, particularly with concept drift.
• Existing approaches struggle with the dynamic and uneven nature of streaming data.

Purpose of the Study:

• To propose a novel framework, DFGW-IS, for classifying imbalanced data streams.
• To effectively handle concept drift and uneven class distributions in streaming data.
• To provide theoretical guarantees and empirical validation for the proposed method.

Main Methods:

• Developed a dynamic feature group weighting framework (DFGW-IS) driven by importance sampling.
• Incorporated a weighted ensemble of sub-classifiers, weighing each by discriminative power and stability.
• Utilized importance sampling to rebalance class distributions within feature groups.

Main Results:

• Derived the theoretical upper bound for the generalization error of the DFGW-IS algorithm.
• Demonstrated significant improvements over competing algorithms on benchmark synthetic and real-world data.
• Achieved superior performance in standard evaluation metrics and parallel running time.

Conclusions:

• The DFGW-IS framework effectively addresses concept-drifting and imbalanced streaming data.
• The proposed method offers a robust solution for complex data stream classification challenges.
• Empirical results confirm the superiority and efficiency of the DFGW-IS approach.