As a team, we currently use a simple mathematical model for planning the size of future releases (in story points). I am looking to overhaul this at the moment so that it more accurately reflects what the team has achieved previously.
Is this a problem that machine learning could be applied to, and if so, where would be a reasonably place to start?
Although, this tool would sell like crazy, it won't be able to handle exceptions and variability (I believe it won't be able to handle vacation period, which is not too complicated for humans). Moreover, some would take its output as something written into stone and not a forecast.
If you are looking for something more scientific or semi-automatic you may want to checkout Troy Magennis' work. He concluded - based on various studies - that cycle times follow a weibull distribution. He also has a spreadsheet that can calculate the probable cycle times.
