Generating Classification Rules According to User's Existing Knowledge

An important problem in applying classification rule induction techniques to practical applications is how to produce rules that are related to the user’s existing knowledge about the domain and his/her current interests. Such rules are interesting to the user, and also easily understood and trusted by the user. They can enhance the existing knowledge of the domain and be relied upon in real-world performance tasks. Past research and applications have shown this to be a crucial requirement in many real-life applications. Existing techniques for dealing with this problem typically use sophisticated methods to bias the rule induction process in order to produce rules that are consistent with the existing knowledge. In this paper, we propose a novel and simple approach. It only needs to pre-process the data using the user’s existing knowledge. It does not make any modification to the rule induction technique. Practical applications have shown that this simple approach is surprisingly effective and flexible. It demonstrates that to obtain useful results, we do not necessarily need to use sophisticated techniques. Sometimes simple approaches may just be sufficient.