A Fine-Tuning Approach to Improve Concept Drift Type Classification Accuracy
Khanh-Tung Nguyen, Quang-Thuy Ha, Xuan-Hieu Phan, Quang-Ngoc Ngo Han · 2024
Concept drift, the phenomenon where the statistical properties of the target variable change over time, presents a significant challenge in maintaining the accuracy of predictive models. Identifying the type of concept drift accurately is crucial for implementing appropriate model adjustments and ensuring robust performance. This paper introduces an enhanced framework for concept drift detection and classification in data streams, building upon the Meta-ADD framework. By incorporating a fine-tuning phase, our approach significantly improves the accuracy of drift-type classification. Experiments demonstrate increased performance in synthetic datasets, confirming the effectiveness of our enhancements.