This paper aims to explain why only a small portion of the students enrolled in a massive open online course (MOOC) actually complete the course. Completion rates in MOOC courses range from ten to 30 percent. The conclusions presented, though, are basically indicators of whether an individual student will drop out. These indicators are: total number of events participated in, days active, whether the student played a video, and the number of textbook chapters explored. An event seems to mean any interaction with the MOOC. On the other hand, the specific course taken and the age of the student were not useful predictors. For some reason, male and female students were analyzed separately, but the analysis process chosen prevented using gender as an indicator.

The analysis is based on three datasets from Harvard and the Massachusetts Institute of Technology (MIT), from 2011 to 2013, encompassing about 1500 students then enrolled in a MOOC. A decision-tree-based algorithm was used to select the indicators. Much of the paper describes the analysis and the choice of a decision tree algorithm. The authors do a good job of mining the data they have, but it appears different data is needed. I found the paper difficult, partly because some basic concepts are not explained. The problem is important, but much more work is needed.