Application of Bayesian Missing Data Model to Industrial Problems

• Activity Number: 179
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Julie Novak*
• Companies: IBM Research
• Keywords: Bayesian Modeling ; Missing Data

Abstract

Missing data is a problem that arises in a wide variety of industries. However, in many cases, these industries may have partial information about the missing components that can be taken advantage of. For example, companies may be interested in understanding whom to send advertisements to in order to obtain the largest number of clients. In this talk, I will first discuss the different types of problems to which this kind of missing data model is applicable. I will then describe how I make use of the hierarchical structure of the model to take advantage of different types of missing information through pooling of information. Finally, both in simulation and with real data, I will demonstrate how this model performs in a retail setting. I compare it to multiple other types of imputation strategies (e.g. nearest-neighbor matching, case-deletion, etc.) and see how different missingness patterns affect its performance over these 'competitor' methods.