I am interested in datasets for knowledge discovery from data streams.
It would be valuable to obtain a real datasets, but artificial are fine for me as well. I am aware of: http://www.liaad.up.pt/kdus/products/datasets-for-concept-drift.
MOA platform (closely related to Weka) can be used to generate datasets with the concept drift, eg. by gradually skewing a hyperplane being the decision boundary. It is also possible to introduce concept drift to existing datasets. Please visit: http://moa.cms.waikato.ac.nz/
For those interested in a repository of data streams with concept drift, I would like to recommend the following link: https://sites.google.com/view/uspdsrepository
Besides the repository, we also conduct a critical analysis of the main challenges in the evaluation of stream learning algorithms using real data in the following paper to be published at DaMi journal (pre-print available at researchgate): "Challenges in Benchmarking Stream Learning Algorithms with Real-world Data".
Preprint Challenges in Benchmarking Stream Learning Algorithms with R...
another idea for controlled concept drift induction onto real world datasets for experimentation and benchmarking: https://github.com/tegjyotsingh/ConceptDriftInduction