Can Machine Learning (or NLP) ever help in estimating effort based on expected lines of code where the model would be trained upon similar applications/files that already exist? If so, is anyone researching this in academia or in any research lab?