Predicting the Fate of Requirements in Embedded Domains

The prediction of the fate of a requirement has a visible impact on the use of the resources associated with it. Starting the implementation of a requirement that does not get inserted in the fi nal product implies that we waste the resources associated with it. Moreover, we also create code and other artifacts that may be dif fi cult to identify and delete once the decision of not to proceed with such requirement has been made, if it will ever been made explicitly instead of simply letting the requirement to fall in the oblivion. Furthermore, an additional require- ment creates an increased complexity in managing the system underdevelopment. Therefore, it would be of immense bene fi t to predict the fate of such requirement as early as possible. Still, such prediction does not seem to be feasible and has not been subject yet to a signi fi cant investigation. In this work, we propose an approach to build such a prediction in a domain in which it is of particular bene fi t — the embedded systems domain, where typically the cost of errors is higher due to its direct impact on hardware. To determine whether a requirement will fail, we consider simply the history of the operations it underwent, treating the requirement as a black box. We use logistic regression to discriminate among the failures. We verify the model on more than 80,000 logs for a development process of over 10 years used in an Italian company operating in an embedded domain. The results are interesting and worth follow-up analysis and extensions to new datasets