New software for predicting students' future academic performance has been created by an Effat researcher. It is powered by machine learning techniques.
Image Source: https://www.pexels.com/photo/books-on-brown-wooden-shelves-3747503/
A prototype has been made that is designed to be easy to use for those without a computer science background. This opens the door for academic counselors and advisors to use this model to support students.
The model takes the GPAs of an undergraduate student’s first three semesters, along with other data about the student, and outputs the predicted final semester GPA with a significant degree of accuracy.
The additional data includes some factors based on the student’s activities in the university, such as assignment marks, quizzes, class tests, or attendance, and some demographic features, including gender, age, family background, and special needs. Gathering this data brought up multiple educational data mining challenges that should be considered in future research.
Results of study
It was found that a Random Forest algorithm predicted the final GPA of the students the best. A Random Forest approach involves taking the results of many decision tree algorithms and averaging them to get a better answer.
This method outperformed the logistic regression and artificial neural network algorithms that were tested.
Benefits of predicting student performance
The benefit of this research can be summarized by the following quote from the study:
“Through the application of machine learning on educational data through prediction, universities and educational institutions will be able to improve teaching and learning outcomes, as well as provide the right support for the different types of students at the institution.”
With predictive models, students who are struggling can get the support from their education institution that will get them on the right track forward. Academic support can be distributed more efficiently to these students at risk. This will decrease the rate of late graduation, dropouts and failure.
Secondary education is important for the functioning and growth of developed societies, so, in the long term, this student risk identification is good for a country as a whole.
Risks of predicting student performance
Predicting student performance using machine learning or neural networks does raise some concerns, including the need to protect the student’s privacy, and ensuring that biases in the model don’t lead to unfair outcomes or discrimination. There is also the risk of errors in the model that could lead to some students getting left behind.
Another concern is how the data is used. Ideally, it would be used to support students who need it. However, an unscrupulous stakeholder might prefer to remove support from students who are predicted to do badly to save money.
Despite these pitfalls to contend with, it is likely that researchers in the area would agree that the benefits outweigh the risks.
Summary
High-quality education is important for the future of our society. In 2015, The United Nations created “The Global Goals”, and goal number 4 is “Ensure inclusive and equitable quality education and promote lifelong learning opportunities for all.”
Using machine learning techniques to predict student performance needs to be thought out carefully, but if used correctly it is a tool that can help us work towards better student outcomes, better education institutions, and a better society as a whole.
Reference
This research was undertaken by Dhekra Alkaf, Faigah Bajammal, and Manal Asrar and was published by Effat University. The full study can be found here: Prediction of Students' Academic Performance using Machine Learning.
