Postboosting Using Extended G-Mean for Online Sequential Multiclass Imbalance Learning.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Neural Networks

Background:

• Online sequential multiclass imbalance learning (OS-MIL) presents challenges with imbalanced class distributions in evolving data streams.
• Existing methods struggle with dynamic data changes and unscaled data.

Purpose of the Study:

• To introduce a novel OS-MIL method, postboosting using extended G-mean (PBG), for neural networks.
• To address limitations of current OS-MIL techniques in handling data scarcity, diversity shifts, and imbalanced distributions.

Main Methods:

• Proposed postboosting using extended G-mean (PBG) algorithm.
• Developed a new update rule for online sequential learning.
• Incorporated dynamic adjustment mechanisms for data scarcity and diversity changes.

Main Results:

• PBG effectively resolves dynamic changing data scarcity (DCDS) and dynamic changing data diversity (DCDD).
• PBG is the only method to address the dense-majority problem and DCDD.
• PBG demonstrates superior performance on unscaled data streams and various imbalanced datasets.

Conclusions:

• PBG significantly outperforms existing OS-MIL methods across multiple challenging aspects.
• The proposed method offers a robust solution for real-world imbalanced data stream learning in neural networks.