MIT uses artificial intelligence to predict online learning drop outs

MIT is trying to figure out which students will thrive in massive open online courses as well as the ones that'll drop out.

MIT said it has begun using artificial intelligence and big data techniques to better predict which students will drop out from open online courses.

The news, which was detailed at a conference on artificial intelligence in education last week, is notable for a few reasons. First, online education is promising, but recent surveys have indicated that there are cultural issues at universities hampering online enrollment. The other issue is that some students simply aren't disciplined enough for online learning.

MIT's techniques touch on that latter point a bit.

Also see: Online education: Higher ed faculty won't buy in

For massive open online courses, MOOCs for short, it's unclear how many people are there to listen to lectures only and what percentage will actually do the homework. Other students may intend to do the homework, but be distracted by other events.

MIT is interested in that latter group that may miss a few deadlines and miss the benefits of the class. These students are deemed stopped out of the class. MIT researchers' predictive model revolved around the following:

  • A set of variables around courses such as time spent per homework problem or time spent on video lectures.
  • Normalized variables compared against class averages.
  • An algorithm that finds correlations between variables and a stopout. The algorithm looks at courses as well as its parts.

MIT's model turned out to be accurate, but researchers also sampled importance based on weightings from similar students in courses as well as new variables. One variable could include time spent on a course on the weekend. That variable indicates motivation as well how busy a person is.