FuzzyBack—A Hybrid Neuro-Fuzzy Ensemble for Concept Drift Adaptation in Stream Mining Using Neural Network

The prevalence of continuous data streams in various domains has posed new challenges in stream mining, particularly with the presence of concept drift. Concept drift is the phrase used to describe changes in the statistical characteristics of data streams over time, rendering initial models ineffective. This research proposes an innovative approach that combines Fuzzy ARTMAP and Backpropagation to address concept drift in stream mining. Fuzzy ARTMAP is a neuro-fuzzy classifier known for adaptability, while backpropagation is a popular training algorithm for neural networks. The approach integrates concept drift detection using Fuzzy ARTMAP, ensemble fusion with backpropagation, and dynamic model updates. By leveraging the strengths of both techniques, the approach aims to enhance the model's learning capability and ensure accurate predictions in the presence of concept drift. The research explores each step of the approach, including experimental setup, evaluation metrics, and comprehensive analysis to validate its effectiveness in addressing concept drift adaptation in stream mining. The ultimate goal is to develop a robust learning framework capable of autonomously adapting to real-time stream mining. The obtained accuracy of our model is nearly 85%.