A Novel Concept Drift Detection Method for Incremental Learning in Nonstationary Environments.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Concept drift poses a significant challenge in dynamic environments, degrading model performance over time.
• Existing methods for concept drift detection often lack efficiency or accuracy in real-world applications.

Purpose of the Study:

• To propose a novel and effective method for concept drift detection.
• To enhance the accuracy of prediction models by adapting to data changes.

Main Methods:

• Development and continuous updating of online sequential extreme learning machines (OS-ELMs).
• Quantification of model modifications due to new data to detect concept drift.
• Validation on synthetic datasets, public real-world datasets, and a wind plant energy prediction problem.

Main Results:

• The proposed method demonstrates superior performance compared to state-of-the-art concept drift detection techniques.
• Detecting concept drift and updating the prediction model leads to improved overall accuracy.
• The method effectively minimizes the number of necessary model updates.

Conclusions:

• The novel OS-ELM-based method offers a robust solution for concept drift detection.
• Adaptive model updating upon drift detection enhances predictive performance and efficiency.
• This approach is valuable for time-series prediction tasks in dynamic environments.