Identifying Vital/Protect Patterns for Classification in Multiple Phenotypes Medical Data

Previous works on medical data only focus on bi-phenotypes medical data. However, with the fast development of medical technique, it is inevitable to classify multiple medical data. In this paper, we first define two patterns (adapting an interestingness measure by statics method) and then propose a new algorithm called MVP that is specially designed to discover such two patterns. At last, applies the discovered optimal rule sets to classify multiple medical data. The key advantage of MVP, as compared to other techniques for pattern discovery, is that MVP directly finds the interesting patterns which are non-redundancy and sense in a specific domain. The experiment results demonstrate the proposed method enables the user to focus on fewer rules and to be assured that the survival rules are all medical domain interesting. The classifier build on the rules generated by our method outperforms existing classifiers