In this paper various ensemble learning methods from machine learning and statistics are considered and applied to the customer choice modeling problem. The application of ensemble learning usually improves the prediction quality of flexible models like decision trees and thus leads to improved predictions. We give experimental results for two real-life marketing datasets using decision trees, ensemble versions of decision trees and the logistic regression model, which is a standard approach for this problem. The ensemble models are found to improve upon individual decision trees and outperform logistic regression.Next, an additive decomposition of the prediction error of a model is considered known as the bias/variance decomposition. A model with a high bias lacks the flexibility to fit the data well. A high variance indicates that a model is instable with respect to different datasets. Decision trees have a high variance component and a low bias component in the prediction error, whereas logistic regression has a high bias component and a low variance component. It is shown that ensemble methods aim at minimizing the variance component in the prediction error while leaving the bias component unaltered. Bias/variance decompositions for all models for both customer choice datasets are given to illustrate these concepts.

No abstract

AND KEYWORDS AbstractTraditionally, recommender systems present recommendations in lists to the user. In contentand knowledge-based recommendation systems these list are often sorted on some notion of similarity with a query, ideal product specification, or sample product. However, a lot of information is lost in this way, since two even similar products can differ from the query on a completely different set of product characteristics. When using a two dimensional, that is, a map-based, representation of the recommendations, it is possible to retain this information. In the map we can then position recommendations that are similar to each other in the same area of the map.Both in science and industry an increasing number of two dimensional graphical interfaces have been introduced over the last years. However, some of them lack a sound scientific foundation, while other approaches are not applicable in a recommendation setting. In our chapter, we will describe a framework, which has a solid scientific foundation (using state-of-the-art statistical models) and is specifically designed to work with e-commerce product catalogs. Basis of the framework is the Product Catalog Map interface based on multidimensional scaling. Also, weshow another type of interface based on nonlinear principal components analysis, which provides an easy way in constraining the space based on specific characteristic values. Then, we discuss some advanced issues. Firstly, we discuss how the product catalog interface can be adapted to better fit the users' notion of importance of attributes using click stream analysis.Secondly, we show an user interface that combines recommendation by proposing with the map based approach. Finally, we show how these methods can be applied to a real e-commerce product catalog of MP3-players.Free Keywords map-based interface, multidimensional scaling, nonlinear principal components analysis, recommender systems, dissimilarity measure Availability The ERIM Report Series is distributed through the following platforms:

Better predictive performance on out-of-sample data can be achieved through an adapted neural network than with various other models

No abstract