Identifying Academically At-Risk Students Based On Personal and Academic Factors Using Machine Learning Techniques

Abstract

Identifying students at risk is crucial for effectively carrying out necessary interventions to ensure students’ successful course completion. Currently, it is difficult to predict academically at-risk students early in the year with great accuracy, so we sought a method to improve the accuracy of estimating these performance trends. This study examined the effectiveness of a Multilayer Perceptron Network in predicting at-risk students throughout the school year. A student performance dataset was tested with different neural network architectures in order to find the highest possible accuracy in determining if a student was at-risk (in the bottom quartile) at the end of the term. This accuracy was then compared to the simple weighted average of student GPA, current grade, and attendance in order to test its effectiveness. We predicted that our network would attain 20% greater accuracy than a simple eye-test method. Our network predicted whether a student would be at-risk among a small dataset with 100 percent accuracy, which was 32 percent greater than the 68 percent accuracy of the eye-test method, supporting our hypothesis. These results are not consistent with recent research on student progress monitoring using Machine Learning algorithms due to the fact that the accuracy was abnormally high, possibly due to the smaller data-set and high-quality data. In the future, we plan on collecting additional datasets in order to obtain more realistic results.