Abstract |
The availability of huge database of customer information has opened up the possibility of using it for purposes of segmentation, targeting, and positioning. Data captured in these databases are non-random and hence excludes the possibility of use of most of the traditional statistical tools for inference. Even the use of simple descriptive tools such as clustering are made difficult by the sheer volume. Over the last two decades computer scientists have worked on developing new algorithms which make application of some simple descriptive multivariate techniques to these databases possible in real time. However, the inferential properties of these techniques are not known at all or are very poorly understood. Bayesian methodology is particularly useful for this purpose since it can incorporate expert opinion which holds the key for making meaningful decisions based on data from these data sets. In this project, we attempt to develop new data mining tools and study their inferential properties from a Bayesian standpoint. |