Customer retention via data mining.pdfVIP

Customer Retention via Data MiningKianSing Ng and Huan Liu (fngkians1, liuhg@.sg)School of Computing, National University of Singapore, 119260Abstract. \Customer Retention is an increasingly pressing issue in todays ever-competitive commercial arena. This is especially relevant and important for salesand services related industries. Motivated by a real-world problem faced by a largecompany, we proposed a solution that integrates the various techniques of datamining, such as feature selection via induction, deviation analysis, and miningmultiple concept-level association rules to form an intuitive and novel approachto gauging customer loyalty and predicting their likelihood of defection. Immediateaction triggered by these \early-warnings resulted from data mining is often thekey to eventual customer retention.Keywords: Customer retention, feature selection, deviation analysis, multiple levelassociation rules, data mining 1. IntroductionIn the last decade, the increased dependency and widespread use ofdatabases in almost every business, scientic and government organiza-tion has led to an explosive growth of data. Instead of being blessed withmore information to aid decision making, the overwhelming amounts ofdata have inevitably resulted in the problem of \information overload-ing but \knowledge starvation, as the human analysts are unableto keep pace to digest the data and turn it into useful knowledgefor application purposes. This situation has motivated some scientistsand researchers in the elds of articial intelligence, machine learning,statistics and databases to put their expertise together to form the eldof knowledge discovery in databases (KDD). KDD seeks to intelligentlyanalyze voluminous amount of information in databases and extractpreviously unknown and useful knowledge (nuggets) from them (Fayyadet al., 1996).Active research in these elds has produced a wide range of ef-fective knowledge discovery techniques like ID3 (Quinlan, 1986) forclassication