Dr. Eduardo Silva Alvarado, director of the Fundación Universitaria Iberoamericana (Iberoamerican University Foundation, FUNIBER), at its headquarters in Guatemala, participates in a study that contrasts various techniques for predicting the dropout of online graduate studies in advanced stages in order to propose a more efficient model.
Currently, the development of online learning poses a great challenge for educational institutions worldwide. Two of these challenges are the lack of knowledge about online learning methodology and the digital divide, as not all students are comfortable with virtual procedures and do not have the same opportunities to access technological tools.
Although the number of students enrolling in online programs has increased, this has also resulted in high dropout and dissatisfaction rates among students, especially compared to studying on a face-to-face campus. This applies mainly to massive open online courses, usually free of charge, known as MOOCs, which can have dropout rates above 80%. Therefore, predicting and retaining students has become a crucial challenge.
Graduation and dropout rates in university studies are considered indicators of the effectiveness of educational institutions and can affect their reputation and even their eligibility for scholarships and government funding. The causes of college dropout are complex and are influenced by various psychological, social, economic and organizational factors.
Dropout can be classified in terms of duration (temporary or permanent) and individual behavior (academic or voluntary). Another way of classifying dropout refers to the time at which it occurs, distinguishing between early, early and late dropout. In this context, it is important to note that this study focuses on voluntary and definitive late dropout during the phase of the development of the Master’s Thesis.
To address this issue, the researchers contrasted the benefits of the optimal probability threshold adjustment technique with other unbalanced data processing techniques, in its application to the prediction of late dropout of graduate students in distance education courses at two universities in the Ibero-American region: the European University of the Atlantic (UNEATLANTICO) and the International Ibero-American University of Mexico (UNINI Mexico), part of the University network in which FUNIBER participates. The ultimate goal is to reduce the gap between traditional models used in face-to-face education and the online learning environment, by applying machine learning tools to make specific decisions. This approach focuses on the technique of optimal probability adjustment to predict graduate university dropout at late or advanced stages. This can be applied independently or in conjunction with other techniques, attributes or algorithms.
According to the methodology used, it was confirmed that a complex consensus model (with threshold of 0.463) and the Random Forests model obtained similar metrics. The adjustment of the optimal probability thresholds in the base models demonstrated the robustness of the Random Forests model in achieving a balance between accuracy (0.56) and recall (0.55), using a threshold close to the default value of 0.5 (0.427). This demonstrates that a base Random Forests model fitted with an optimal threshold provides robust results and avoids the need to use data alteration techniques that could introduce errors.
Significant variables were found to be primarily related to academic environment, and variables with an explicit time component, such as length of coursework, length of academic status, and extensions, were found to have the greatest significance in differentiating between different classes in the context of late dropout. Social and demographic variables did not seem to influence this situation as much.
The results suggest that the approach of fitting the optimal probability threshold provides better generalization of the late dropout prediction model compared to other data resampling techniques. However, more variables should be considered and it is recommended to expand the sampled population, use additional resources such as Python, and improve the machine learning algorithms in future research.
This innovative study marks a milestone in the academic field, generating hope for universities to intervene in a timely manner and provide adequate support to students who may be at risk of dropping out of their studies, mainly in their last phase of study. In this way, late-stage dropout rates are minimized.
If you want to know more about this study, click here.
To read more research, consult the UNEATLANTICO repository. The Iberoamerican University Foundation (FUNIBER) offers the Master’s Degree in Education. A program oriented to the development of digital competencies of teachers of the 21st century and the application of technology to education.