Research Shows Sharply Different Learning Behaviours On The Part of Introductory Level Programming Students
New research by academics in DCU’s School of Computing , just published in the prestigious journal Future Generation Computer Systems, indicates the presence of sharply different learning behaviours on the part of introductory level programming students. This research showed that those who perform well in the terminal exam tend to engage in far more practice with programming course material when compared to their less well performing counterparts, whose learning behaviour consists mainly of re-reading lecture notes. The research also discovered that mathematical modelling can be used to detect groups of learners, with similar behaviours, early in the course. This modelling can then be used to predict outcomes in the initial stages of the programme of study .
The study used data recorded for students on two programming-related courses during two academic years (2017/2018 and 2018/2019). The difference between the two courses related to the level of computing knowledge and task requirements, allowing investigation of the performance of students at different levels of programming learning. Critically, the courses were delivered in a combination of conventional and online formats In particular, students have physically attended the lecture sessions in lecture halls and have conducted all learning activities on a bespoke online system.
Sophisticated modelling techniques were used by the authors to detect the presence of different cohorts of students on their respective courses. School of Computing Postgraduate Student Tai Tan Mai, supervised by Prof Martin Crane and Dr. Marija Bezbradica, found evidence that students from different cohorts, corresponding to different patterns of learning behaviours, showed distinct levels of performance in the exams at the end of the module. Having detected this, the authors showed that the interactions of such groups with items of course material (such as lecture notes), varied markedly. In order to predict, at the early stages of the study period, whether the students would eventually either pass or fail the final module exam, the authors applied a set of machine learning algorithms to weekly learning data. In the study, novel methods have also been developed to pre-process the dataset, improving the overall performance of the analysis and prediction.
The study was facilitated by access to data from the course dashboard, Einstein, a bespoke Virtual Learning Environment (VLE), developed in-house by Dr. Stephen Blott, for use in the School of Computing, DCU. The learning data was generated by participants on the programming courses. Such modern VLEs provide the ability to automatically record a large amount of data at fine-grained levels of interaction with the course material. This research has shown the potential of the interaction data to improve the pedagogical value of online teaching and learning and has proven hugely beneficial, during COVID-19, in maintaining the quality of programming education in DCU.
The research is supported by the Irish Research Council and ADAPT Research Centre. The paper can be accessed via: https://doi.org/10.1016/j.future.2021.08.026