A change prediction model for embedded software applications

Abstract

A key aspect of effective software development is the ability to quantify and predict software product quality. Software quality is the degree to which software possesses desired attributes, including portability, reliability, testability and maintainability. Insofar as software with high change traffic affects its maintainability and reliability, a model which produces a change-traffic predictor metric may be useful. The information such a model would provide could be used to help estimate the development cost and effort. Resources could be better allocated to those areas where additional attention may be required. Software changes normally occur due to new requirements or errors in the software, and so a change-traffic metric is not necessarily a good proxy for errors. Users should define their thresholds and ranges of acceptability. This paper identifies metrics collected from embedded Ada software that had a correlation with the change traffic of that software. Using multiple linear regression analysis and sample data from up to 287 embedded Ada software modules, change prediction models yielded values for the average absolute difference between predicted and actual changes per module of less than 3, and an adjusted-R/sup 2/ value of 0.57 for the full sample.