A Diversity Framework for Dealing With Multiple Types of Concept Drift Based on Clustering in the Model Space

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Data stream applications face challenges with various concept drift types.
• Existing methods often struggle with multiple drift types, limiting predictive performance.

Purpose of the Study:

• To propose a novel diversity framework for handling multiple concept drift types in data streams.
• To leverage ensemble diversity for improved adaptation and handling of recurring concepts.

Main Methods:

• Utilizing a diversity framework with ensemble methods.
• Employing clustering in model space to build diverse ensembles.
• Identifying recurring concepts through ensemble diversity.

Main Results:

• The diversity framework demonstrated comparable or superior prequential accuracy against existing approaches.
• Effectiveness was particularly noted in scenarios with recurring concepts or concept similarity.
• Experiments were conducted on 20 synthetic and three real-world data streams.

Conclusions:

• The proposed diversity framework offers a robust solution for concept drift in data streams.
• Diversity in ensembles accelerates adaptation to various drift types, especially recurring or similar ones.
• This approach enhances overall predictive performance in dynamic data environments.