Graph ensemble boosting for imbalanced noisy graph stream classification.

Area of Science:

• Machine Learning
• Data Mining
• Graph Analytics

Background:

• Stream data with structural dependencies and imbalanced class distributions present significant challenges for classification models.
• Noise in imbalanced data can be misidentified as minority samples, leading to reduced model accuracy.
• Concept drift in graph streams further complicates accurate classification.

Purpose of the Study:

• To propose a novel classification model, graph ensemble boosting, for imbalanced graph streams containing noise.
• To address the challenges of class imbalance, noise, and concept drift in graph stream classification.
• To improve the accuracy of identifying minority class samples in dynamic graph data.

Main Methods:

• An ensemble-based framework partitions graph streams into manageable chunks.
• A boosting algorithm unifies discriminative subgraph pattern selection and model learning within each chunk.
• An instance-level weighting mechanism dynamically adjusts sample weights to handle concept drift and emphasize difficult samples.

Main Results:

• The proposed graph ensemble boosting method demonstrates clear benefits in handling imbalanced and noisy graph streams.
• Experimental results on real-life imbalanced graph streams validate the effectiveness of the boosting design.
• The approach successfully tackles both class imbalance and concept drift in graph stream classification.

Conclusions:

• Graph ensemble boosting provides an effective solution for classifying imbalanced graph streams with noise.
• The method enhances the ability of learning models to accurately identify minority samples in dynamic graph data.
• The unified framework for subgraph pattern selection and model learning, combined with dynamic weighting, is crucial for performance.