Reacting to different types of concept drift: the Accuracy Updated Ensemble algorithm.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Data stream mining is crucial for real-time applications like banking and telecommunications.
• Concept drift, unpredictable changes in data distribution, poses a significant challenge for stream learning algorithms.
• Existing methods often specialize in specific types of concept drift, limiting their general applicability.

Purpose of the Study:

• To develop a novel data stream classifier, the Accuracy Updated Ensemble (AUE2), capable of adapting to various types of concept drift.
• To improve upon existing methods by combining ensemble accuracy weighting with incremental learning techniques.

Main Methods:

• AUE2 integrates accuracy-based weighting from block-based ensembles with the incremental learning of Hoeffding Trees.
• The proposed AUE2 algorithm was experimentally evaluated against 11 state-of-the-art stream classification methods.
• Performance was assessed across diverse concept drift scenarios, including gradual, abrupt, and recurring changes.

Main Results:

• AUE2 demonstrated the highest average classification accuracy among all compared algorithms.
• The AUE2 classifier proved to be more memory-efficient than other ensemble approaches.
• AUE2 showed robust performance across various drift types and in static environments.

Conclusions:

• AUE2 is a highly effective and memory-efficient classifier for data streams experiencing concept drift.
• The algorithm's ability to handle diverse drift types makes it suitable for a wide range of real-world applications.
• AUE2 represents a significant advancement in adaptive stream learning, offering improved performance and resource utilization.