Mining the most interesting patterns from multiple phenotypes medical data

Mining the most interesting patterns from multiple phenotypes medical data poses a great challenge for previous work, which only focuses on bi-phenotypes (such as abnormal vs. normal) medical data. Association rule mining can be applied to analyze such dataset, whereas most rules generated are either redundancy or no sense. In this paper, we define two interesting patterns, namely VP (an acronym for "Vital Pattern") and PP (an acronym for "Protect Pattern"), based on a statistical metric. We also propose a new algorithm called MVP that is specially designed to discover such two patterns from multiple phenotypes medical data. The algorithm generates useful rules for medical researchers, from which a clearly causal graph can be induced. The experiment results demonstrate that the proposed method enables the user to focus on fewer rules and assures that the survival rules are all interesting from the viewpoint of medical domain. The classifier build on the rules generated by our method outperforms existing classifiers.