Data Mining Boilermakers

Purdue University is applying basic data mining principles to identify at-risk students based on criteria culled from their existing course management system(s) (CMSs). The concept of data mining isn't exactly new, but Purdue's approach to identify and intervene with students who may be close to failure or dropping out is a new spin on the old practice.

Capturing and analyzing such data as time-on-task for online course components is starting to point to trends that can then be managed and predicted. Student behavior and interaction with a CMS can be measured, and even qualified to a certain extent. This information can then be parsed to create a model that anticipates performance based on the individual student's behavior. It can, ostensibly, also be rolled up to more complex models studying group behavior, by course, by instructor, by time of day, or shirt size such that macro trends can also be exposed. Perhaps unsurprisingly, it's worth noting that the majority of major CMSs can capture and report data that might be lying dormant today. Props to Purdue for noticing, and for recongnizing the potential. It's entirely possible that other schools are doing this already, and this one just happened to pick up the ears of a news site. If that's the case, or if your school(s) is (are) doing something along these lines, please comment the post and let's get some ideas exchanged. Naturally, the predictive ability of the models is dependent on the quality and variety of data fed in, but still ... this is pretty cool. Check out the brief article here, at eSchool News.