Data Mining for Learning Analytics: does lack of engagement always mean what we think it does?

Context and Objectives Learning Analytics (LA) has the potential to utilise student data to further the advancement of a personalized, supportive system of HE (Johnson et al., 2013). A number of LA systems are now being developed but there have been few studies that have analysed the usage of Virtual Learning Environments (VLE) in order to identify which analytics techniques and sources of data accurately reflect student engagement and achievement. Methods The interactions of 66 students with a Level 4 programming module on a VLE have been analysed via the simple K-means clustering algorithm to identify classes of behaviour and their characteristics. Results Two prominent classes were found with students achieving higher marks attending the lectures and tutorials more regularly and accessing all types of material on the VLE more frequently than students in the lower achieving cluster. However, there were a number of exceptions that had low levels of engagement that gained high marks and vice versa. Discussion A student’s prior experience and characteristics of their degree programme need to be taken into account to avoid incorrectly interpreting high and low levels of engagement. Conclusions The number of times students view online module materials will be an important factor for inclusion in any predictive LA models but must be able to take into account the differences in student backgrounds, delivery styles and subject