Benchmark Data-sets for Concept Drift where important predictors (independent variables) change with time or stream of observations I'm currently searching the web and literature for streaming classification datasets with concept drift.  I've found a number of synthetic datasets where over time the important predictors either change in their "predictive" nature.
For example here is a paper with a synthetic data-set with four blocks of step style concept drift.  
I've also been checking out the references on the wiki article on concept drift.
My question is.. what other REAL streaming datasets out there express concept drift in classification problems.  In particular I'm looking for datasets where the set of important features completely changes with time.  I've already simulated my own dataset which I vaguely discuss here.  Any help is appreciated.
 A: The UCI database is always a good place to start:
http://archive.ics.uci.edu/ml/datasets.html?format=&task=cla&att=&area=&numAtt=&numIns=&type=ts&sort=nameUp&view=table
Weather would also be interesting.  If you picked a location with known seasonal variances in weather patterns, you may be able to predict something from that.
A: A good option for changing feature relevance could be a textual data stream, where features are related to words. For example, sentiment classification in tweets.
Unfortunately, researchers are only allowed collect data for own use via Twitter api, but not distribute ready-made datasets publicly, I am not available of any publicly available set of such kind. 
A: You can check this datasets on MOA framework page. It's good for start. There is some more in web from various competitions.
A: I ended up using the simulated hyperplane dataset but used the dimension weights over time as a measure of true feature importance
Code to generate hyperplane data and dimension weights
My Paper
