Probabilistic rule induction from a medical research study database.

This paper describes and applies the ITRULE algorithm, a generalized rule induction algorithm of Smyth and Goodman, for the exploratory analysis of a subset of data from the National Heart, Lung, and Blood Institute Growth and Health Study. The ITRULE algorithm is used as an effective, parsimonious method for detecting informative relationships in a large set of study variables. The potential complexity of the search among many variables is controlled by applying an information theoretic rule preference measure, the J-measure, to limit rule specialization.