Introduction to special issue on machine learning for business applications

In recent years we have witnessed a dramatic increase in novel uses of machine learning for business applications. These applications cover a wide range of traditional as well as new business activities from intelligent customer segmentation for direct marketing to intelligent stock market analysis and the analysis of the long tail in the new Web-based economy. Innovative machine-learning techniques that make use of different aspects of newly emerged business data, such as customer feedback and behavior data and social network data are under development. To further foster the advances of the most recent and exciting work on machine learning for business applications, we initiated this special issue. The response to our call for papers was very strong. After careful reviews of the submissions by international experts in the area, we have selected six articles for this special issue. Since the theme of the special issue is business applications, which relates to money, three of the six articles are about applying machine learning and data mining to the financial market. In " Prediction in Financial Markets: The Case for Small Disjuncts " , Vasant Dhar proposes that a counterintuitive method of learning many small disjuncts instead of a single model can provide a credible model for financial market prediction, a problem with a high degree of noise. His results suggest that for problems characterized by a high degree of noise and lack of a stable target concept, which are widespread in many real-world applications, constructing sets of small rules and then reconstructing them periodically is a promising approach. In " A Learning-Based Contrarian Trading Strategy via a Dual-Classifier Model " , Szu-Hao Huang et al. first discuss a return anomaly proposed by behavioral financial experts called overreaction. Overreaction means that the past winning and losing stocks will have return reversal after a holding period of certain lengths. This phenomenon can be exploited in automatic trading approaches including the contrarian trading strategy. The authors develop a novel dual-classification method for the con-trarian trading strategy to improve the reliability of predicting stocks with the return reversal to form better portfolios. Various experiments showed that their system can increase the returns of portfolios impressively. However, their method does not consider the transaction cost in stock trading. In " CORN : Correlation-Driven Nonparametric Learning Approach for Portfolio Selection " , Li et al. present a statistical correlation-based method named CORN to select the best stocks …