Most people know that a student who does well in a class would likely do well in another class, but could you predict whether a student would succeed based on how many times they logged into the campus network?
With the explosion of data sets now available to them, Purdue University was able to analyze data from hundreds of thousands of successful graduates, to help prediction models for graduation rates and on-time graduations. Gerry McCartney, VP & CIO at Purdue, spoke about these new data sets and how Purdue usee them at the recent AGENDA16 conference.
With more data available (from things like network login records and geolocation), they were able to make better predictions. Purdue then works with existing students and their advisors to help avoid any stumbling blocks towards graduation. One big surprise was how data science has turned traditional analytics on its head, McCartney said.
“This is the way that data science is different from conventional models. The way we were taught … is that you have to develop a hypothesis first, then you test it using statistics. This works exactly the other way around - here’s the data, the data is going to tell you the story. Traditional scientists will say ‘Well, that’s wrong - that’s fishing.’ It’s not fishing, because this is all the data - there is no other data. This is it. All of reality is right here for this topic. How you interpret it now - that’s super important.”
Register now to listen to the complete audio from McCartney’s session.
To continue reading this article register now