Maximizing the predictive value of production rules

摘要

A new method for empirical rule induction under conditions of uncertainty is described. The problem is to find the single best production rule of a fixed length for classification. Predictive value maximization (PVM), a heuristic search procedure through the hypothesis space of conjunctions and disjunctions of variables and their cutoff values, is outlined. Examples are taken from laboratory medicine, where the goal is to find the best combination of tests for making a diagnosis. Resampling techniques for estimating error rates are integrated into the PVM procedure for rule induction. Excellent results for PVM are reported on data sets previously analyzed in the AI literature using alternative classification techniques.