A Drift Detection framework that uses historical information to recognize concept drifts in the past and uses that information to bolster the training set.
If you use this repository in publishing your findings, please cite this work
@inproceedings{fellicious2024driftgan,
title={DriftGAN: Using historical data for Unsupervised Recurring Drift Detection},
author={Fellicious, Christofer and Julka, Sahib and Wendlinger, Lorenz and Granitzer, Michael},
booktitle={Proceedings of the 39th ACM/SIGAPP Symposium on Applied Computing},
pages={368--369},
year={2024}
}
We have the preprint on Arxiv at https://arxiv.org/abs/2407.06543
Datasets can be downloaded at https://github.com/ogozuacik/concept-drift-datasets-scikit-multiflow