Predicting the time-to-deliver of software changes

In this thesis we examine the application of survival analysis on time-to-deliver data. Successful prediction of the time necessary to deliver a new feature or fix a reported defect can assist in various phases and aspects of software development. We identify and try to overcome limitations when dealing with time-to-event data. Our proposed methodological framework includes use of survival analysis, utilization of incomplete information that might be available as censored data, and incorporation of random-effects through mixed-effects models for identification of hierarchical/clustered data within our dataset. We explore and experiment with a dataset from a large scale commercial software over a twelve year period of time. We show that we can successfully implement survival analysis, and that incorporation of random-effects provides a considerable advantage, however, incorporation of censored information is not proven to be advantageous in this case.