Lately researchers have been using edX data to really delve into exactly what happens in MOOCs. For example, just a couple of weeks ago, MIT, Harvard and Tsinghua University (China) researchers discovered (to the surprise of many) that no matter where they start from in terms of previous formal education, MOOC students all show relatively equal learning gains.
In a new study, MIT researchers Kalyan Veeramachaneni and Una-May O’Reilly explore what type of students are most likely drop out of MOOCs, using machine learning models to predict dropout behavior.
The researchers divided students from the first edX course, Circuits and Electronics, into five groups:
• No attempts: never submitted an assignment
• Discussion generators: participated in forums
• Content generators: edited wikis
• Fully collaborative: both participated in forums and edited wikis
• Passive collaborator: submitted assignments, but didn’t participate in forums or edit wikis
For the study, they attempted to predict which students out of the four active cohorts (excluding “No attempts”) would drop out on a week-by-week basis.
Here’s what they found:
• After the first week, they were able to predict drop-outs pretty well.
• The most influential predictor was the “pre-deadline submission time,” which is the amount of time between when a student starts to work on the weekly problem set and when the assignment is due.
• A student’s average number of weekly submissions relative to other students is a good predictor of whether the student will stay in the course, as is the lab grade each week and the average length of discussion posts.
• For each week, accurate predictions can be made based on data from the previous four weeks. It is easy to predict who will drop out one week in advance.
The researchers are continuing to investigate how this information might be used to help students at risk of dropping out. In the meantime, the main suggestion that jumps out is encouraging the students to start assignments further in advance of the deadline, perhaps by sending emails reminding them that assignments are coming up and estimating how long each assignment might take.