A systematic literature review: Recent techniques of predicting STEM stream students

Texto completo

Nowadays, fewer students are choosing to enroll in STEM (science, technology, engineering, and mathematics) fields. STEM students in schools and in higher educational institutions appear to be waning, as evidenced by low secondary school STEM enrolments. To add to this, there are also STEM stream students who dropped out and switched to non-STEM streams. This resulted in a shortage of qualified candidates for STEM-based higher education programmes, and subsequently an insufficient number of STEM graduates. Researchers have found several potential contributing factors that may have impacted students’ selection of STEM. However, this relationship is still unclear and needs further investigation. This goal of this systematic review is to assess the factors that can be used to predict students’ selection of a STEM major using existing techniques. To do this, PRISMA’s Systematic Literature Review (SLR) process was used to map the findings of previous studies based on the designed research questions (RQs). More specifically, the objective of this analysis was to compile, summarise, and assess related works in order to identify current contributing variables, potential techniques, dataset characteristics, challenges, and future directions within the scope of this investigation. Papers published in major online scientific databases, including Science Direct, Scopus, IEEE Xplore, ACM, ProQuest, and Springer, between 2011 and April 2021 were identified and analyzed. Although there were 1248 publications found through extensive SLR selection processes using specific inclusion and exclusion criteria, only 121 articles were selected. After being analyzed, only 16 articles were found to have discussed about machine learning (ML) techniques and showed that the most accurate predictions were possible based on different variables or factors. In addition, the dataset characteristics were found to have impacted the accuracy of the prediction results. However, the available evidences were limited, and the output findings from each study reviewed were relatively diverse. Therefore, evidences discussing the potential usefulness of ML techniques to analyse the relationship between contributing factors should be strengthened.