Exposure model for prediction of number of customer reported defects

Abstract

The paper describes a mathematical model, the Exposure Model, for the prediction of customer reported defects in a large software system. In the model, exposure is defined as the likely fraction of the original defects in the software system that is reported by any customer, in any month and against any version of the system. The basic idea is to try to model exposure as the product of the isolated effects of a few customer- and system-characteristics. The model has been used for several purposes: to better understand defect detection mechanisms, and to better predict the resources required for defect correction. Also, the model has enabled us to make an early estimate of product quality, and thereby give valuable input for other important purposes.