Student retention is of great concern to higher education institutions, and a higher dropout rate may be reflected in the university’s reputation. Extensive studies have attempted to identify at-risk students in an attempt to reduce student attrition as much as possible.
In an attempt to predict student performance, the authors present a predictive model based on feed-forward multilayer neural networks that includes “academic, demographic, institutional, psychological, and economic factors.” The bulk of the dataset used consisted of traditional students (18–22 years old) at a British university.
The authors carefully normalized input and output vectors. The model was implemented using MATLAB. The end-of-year results provide statistically valid results. The model reflects the United Kingdom’s educational system; it may not apply to other countries, where degrees are not classified as first or upper second. In addition, the classifiers used may not apply to nontraditional students, such as older and/or working adults, students with family members, and those who are economically independent. In these cases, the approach can be tailored to replace the classifiers with more relevant ones, for example, classifiers based on a particular degree structure.
A psychological motivation factor measured the sustained level of activity throughout the course. This is an example of a factor that may be relevant to traditional students, but not nontraditional ones. For nontraditional students, family emergencies may skew this measure at times, especially if students are single parents, for example.
Researchers interested in identifying at-risk higher education students should read this paper.