Method of approximate logistic regression analysis based on discovery of itemsets with statistical background

Abstract

A method for discovering itemsets with statistical background (ItemSB) between two variables of interest has been proposed; applying the method for discovering ItemSB with a background of correlation between two variables, we propose an approximate method of logistic regression analysis based on the discovery of small groups with a certain combination of attributes. Since multiple logistic curves are discovered by this method, we examined analytical and prediction methods using these curves. Experimental results of applying this method to publicly available medical data are reported.