Regression methods are used to model software fault detection effectiveness in terms of several product and testing process measures. The relative importance of these product/process measures for predicting fault detection effectiveness is assessed for a specific data set. A substantial family of models is considered, specifically, the family of quadratic response surface models with two way interaction. Model selection is based on "leave one out at a time" cross validation using the predicted residual sum of squares (PRESS) criterion. Prediction intervals for fault detection effectiveness are used to generate prediction intervals for the number of residual faults conditioned on the observed number of discovered faults. High levels of assurance about measures like fault detection effectiveness (residual faults) require more than just high (low) predicted values, they also require that the prediction intervals have high lower (low upper) bounds.
展开▼
