Knowledge maintenance on data streams with concept drifting

In this paper, we propose a method of hiding sensitive classification rules from data mining algorithms. Our idea is to reconstruct a dataset according to the classification rules that have been checked and agreed by the data owner for release to data sharing. Unlike other heuristic approaches, firstly, our method classifies a given dataset. Then, a set of classification rules are shown to the user. User then identifies the rules that are to be hidden. After that we generate a new decision tree that has only non-sensitive rules. A new dataset can then be reconstructed with no more and no fewer classification rules that can be derived. Our experiments show that this approach is efficient and effective