Creating an Educational Roadmap for Engineering Students via an Optimal and Iterative Yearly Regression Tree using Data Mining

Abstract

Targeting high academic standards is required in engineering studies. Advisors usually play an important role in helping students keeping good records and transcripts along their educational path by helping them choose their courses and keeping track of their grades. However, performance in some courses in the curriculum embodies determining repercussions and might inadvertently jeopardize the overall students' Grade Point Average (GPA) in an irreversible manner. The purpose of this paper is to draw an educational roadmap that helps advisors and students being aware of the turning points that decisively affect their overall cumulative GPA and act upon a current outcome. This roadmap is based on Classification and Regression Trees where nodes and branches denote the aforementioned courses and students' performance, respectively, with the ultimate outcome being the overall student's GPA upon graduation. The tree is constructed based on a relatively large number of records with 10-fold cross-validation and testing. Moreover, the tree is produced on a yearly basis with a twofold objective. The first is to secure a high level of precision by applying it over a short period of time and the second is to allow for injecting each-year computed GPA with the remaining courses as to reflect the actual situation with maximum vraisemblance. This iterative and recursive tree achieves a very close tracking of students' performance and provide a powerful tool to rectify courses' track and grades for each student individually while aiming at a predefined final GPA. Furthermore, the choice of the optimal tree was carefully examined in the light of the relatively elevated number of attributes. In this context, diverse models were created and performance and/or precision were computed in terms of different values of "pruning levels" and "splitting criteria". The choice of the best tree to be adopted for advising is thoroughly explained. Besides, it is shown, in this context, that the structure of the tree remains highly versatile in the sense that it can be revisited at any point for further assessment, adjustment, and expansion. Finally, yet importantly, simulation results were carried out using Matlab CART and demonstrated high efficiency and reasonably precise results.