TS-DM: A Time Segmentation-Based Data Stream Learning Method for Concept Drift Adaptation.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Concept drift is a common challenge in data streams due to evolving data distributions.
• Existing methods often struggle with optimally managing data samples during concept drift, leading to potential knowledge loss and reduced model accuracy.

Purpose of the Study:

• To propose a novel time segmentation-based data stream learning method (TS-DM) for effective concept drift adaptation.
• To enhance the generalization and robustness of machine learning models operating on streaming data.

Main Methods:

• Developed a chunk-based segmentation strategy to differentiate normal and drift data chunks.
• Introduced a chunk-based evolving segmentation (CES) strategy to mine and segment data where old and new concepts coexist.
• Implemented a warning level data segmentation process (CES-W) and a high-low-drift tradeoff handling process.

Main Results:

• Experimental evaluation on synthetic and real-world datasets demonstrated the efficiency of the TS-DM method.
• The proposed method showed superior performance compared to several state-of-the-art data stream learning techniques.
• TS-DM effectively addresses the challenge of improperly keeping or discarding data samples during concept drift.

Conclusions:

• The TS-DM method offers an efficient and robust approach to concept drift adaptation in data streams.
• The proposed segmentation and tradeoff strategies enhance the model's ability to learn from evolving data.
• This work contributes to improving the accuracy and reliability of machine learning models in dynamic environments.