Challenges on Classifying Data Streams with Concept Drift
Eduardo Victor Lima Barboza, Paulo Ricardo Lisboa de Almeida · 2022
Concept Drift is a common problem when we are working with Machine Learning. It refers to changes in the target concept over time, which may deteriorate the model’s accuracy. A recurrent problem on concept drift is to find datasets that reflect real-world scenarios. In this work, we show some datasets known to have Concept Drift, and propose changes in an existing method (Dynse), which include making it capable of handling data streams, instead of batches, and putting some trigger on it, to make its window adaptive by detecting concept drift.