Itemset-based mining of constraints for enacting smart environments

In order to automatically control the environment, smart systems should have sufficient rules, which describe expected system's behavior. While such rules may be added man-ually, usually this requires considerable efforts, often surpassing those that users are willing to spend to setup the system. In this paper, we propose a novel technique to mine such rules automatically, given a sensor log from the environment. In particular, we mine itemsets, but we consider abnormal drops in the frequency of variable state combinations w.r.t. the frequency of their subsets, which represent undesirability of these combinations. We evaluate the technique both on simulated and real datasets, showing that the approach is effective and promising for further extensions.