Tackling Virtual and Real Concept Drifts: An Adaptive Gaussian Mixture Model Approach

Real-world applications have been dealing with large amounts of data that arrive over time and generally present changes in their underlying joint probability distribution, i.e., concept drift. Concept drift can be subdivided into two types: virtual drift, which affects the unconditional distribution p(x), real conditional p(y|x). Existing works focuses on However, strategies to cope may not best suited for since class boundaries remain unchanged. We provide first depth analysis differences between impact drifts classifiers' suitability. propose an approach handle both called On-line Gaussian Mixture Model With Noise Filter For Handling Virtual Real Drifts (OGMMF-VRD). Experiments seven synthetics real-world datasets show OGMMF-VRD outperforms other approaches separate mechanisms deal drifts. It also has more stable rankings smaller drops performance during drifting periods than existing ensemble approaches, thus being reliable adoption practice.