Brute-force mining of high-confidence classification rules
Jr. Roberto J. Bayardo · 1997
This paper investigates a brute-force technique for mining classification rules from large data sets. We employ an association rule miner enhanced with new prun ing strategies to control combinatorial explosion in the number of candidates counted with each database pass. The approach effectively and efficiently extracts high confidence classification rules that apply to most if not all of the data in several classification benchmarks. Introduction Several data mining tasks require dividing up the entities of a database into various classes. Junk-mailers are wellknown users of classification technology, using it to avoid sending out flyers to persons unlikely to be interested in the product being promoted. The task requires a classifier that is usually automatically generated from a "training database " of pre-classified entities. Several approaches have appeared in the AI, statistics, and data-mining literature, and some methods made to scale to large data sets [Shafer et al. 96]. B...