Pattern-Based Target Selection Applied to Fund Raising

Abstract

This paper proposes a new algorithm for target selection. Thisalgorithm collects all frequent patterns (equivalent to frequent itemsets) in a training set. These patterns are stored e?ciently using acompact data structure called a trie. For each pattern the relativefrequency of the target class is determined. Target selection is achievedby matching the candidate records with the patterns in the trie. Ascore for each record results from this matching process, based uponthe frequency values in the trie. The records with the best score valuesare selected. We have applied the new algorithm to a large data setcontaining the results of a number of mailing campaigns by a Dutchcharity organization. Our algorithm turns out to be competitive withlogistic regression and superior to CHAID.

Keywords: frequent item sets, target selection, datamining, frequent patterns

JEL Classification: R4, M, M11, C89