Machine Learning For Direct-Marketing Response Models

The last decade has witnessed phenomenal growth in direct marketing around the world. According to the Direct Marketing Association, revenues generated through direct marketing channels in the U.S. alone reached US$1.7 trillion in 2000, accounting for a significant portion of the country's economy activity. In the year 2005, direct marketing was legislated in China. It is reported by China Economic Times that the total business volume of direct marketing in China was about ￥35 billion in 2004. According to some experts may be concerned, focus on China direct marketing, the growth rate is expected to reach 20% or even higher. In the more open year 2006, the total volume may reach ￥50 billion.The expanding of direct marketing brings in new challenges focus on quantify research. Take catalog companies for example. Data analysts in direct marketing seek models to identify the most promising individuals to mail to and thus maximize returns from solicitations. Given budgetary limitations, typically a fraction of the total customer database is selected for mailing. Therefore, a key task in the process is the development of models to identify the most promising individuals, which brings in the topic of response prediction. Since the predictive performance of consumer response models has a direct impact on the effectiveness and profitability of direct marketing operations, accurate prediction of consumer response has become a top priority for many business.Until recently, statistical methods have dominated the modeling of consumer responses to direct marketing, such as logistic regression and discriminate analysis. Besides, machine learning methods have been adopted to solve classification problems such as predicting bankruptcy and modeling consumer choice. Some methods used frequently are classification and regression trees (CART) and artificial...