A Unified Framework for Detecting Gradual and Abrupt Concept Drifts in Data Stream Mining: The Concept Drift Detection Framework with Hybrid Meta-Learning (CDDF-HML)
G. V. S. N. R. V. Prasad, Kapil Dev Sharma · International Journal of Electrical and Electronics Engineering · 2024
The dynamic structure of data streams provides major challenges for sustaining prediction model accuracy over time. Concept drift, defined as changes in underlying data distributions, has been proven to have a considerable impact on the performance of machine learning models in real-time applications. While earlier methods often focus on either slow or abrupt concept drifts, a unified framework capable of identifying both types quickly is absent. As a result, to overcome the issue mentioned above, we propose the Concept Drift Detection Framework with Hybrid Meta-Learning, abbreviated as CDDF-HML. This incandescent method applies meta-learning, adaptive feature selection and ensemble-based process to address both slow as well as sudden concept drifts. Due to this, the framework is most appropriate in dynamic data stream mining, where the underlying structure is continually changing. It showcases how it can identify deviations of ideas with further capability in accommodating various data conditions. The study also performs the comparative analysis with other techniques to demonstrate that CDDF-HML is really an effective tool for discovering concept drift. The future possibilities of CDDF-HML include the implementation of the method within specific domains, further development of granular adjustment approaches, structural and extensional amendments to scalability, and partnerships with professionals from various industries. It is beneficial in the improvement of the concept drift detection in data stream mining so that the reliability of the model can be assured in dynamic data situations.