Exceptional Model Mining: a logistic regression model on cancer registry data.

Finding interesting patterns in a cancer registry data can provide oncologists and medical experts with a new perspective to improve healthcare for cancer patients. In this paper, we apply a supervised local pattern mining called Exceptional Model Mining. Its aim is to find subgroups in the data that somehow behave differently from the norm. This behaviour is captured by a model and the interestingness of a subgroup is assessed according to a quality measure. In particular, we develop a logistic regression model and propose a few different quality measures based on statistical tests and probability distributions. Additionally, we provide a statistical test, the permutation test, to assess the interestingness of a subgroup from a statistical point of view. The results of the experiments show that the proposed model can retrieve some subgroups that may be of interest for doctors. Moreover, the results of the permutation test show that the most interesting subgroups are also statistically significant.