Predictive Analytics for Student Academic Performance Using Educational Data

Abstract Educational data mining (EDM) techniques, as used in learning analytics, have been widely embraced as means of predicting students' academic performance. Machine learning models that can predict such parameters as course grade, grade point average, and likelihood of dropping out can be generated using demographic, previous academic record, and LMS engagement information. Recent surveys and studies suggest that compared to conventional statistical approaches, machine learning methods, particularly the use of random forest, gradient boosting, support vector machine, and neural networks, show better results if provided with data from diverse sources. At the same time, systematic reviews reveal methodological problems, such as heterogeneity of approaches, weak external validity, and ethical issues not being addressed properly. This paper suggests a framework for predicting end-of-semester academic performance in higher education institutions using institutional data.[1] It describes a quantitative non-experimental design where supervised machine learning models are trained using demographic, prior achievement, and LMS engagement features with careful preprocessing, cross-validation, and comprehensive metric assessment. The study will be aimed at (a) producing accurate predictions of students' academic performance, (b) identifying predictors that will help to develop strategies of teaching and supporting students, and (c) demonstrating how the predictive model will serve as an engine for early warning systems and intervention.[2] Combining recent research findings on EDM and learning analytics, the paper provides both the methodology and the conceptual basis for the implementation of predictive analytics in higher education.[3] Searched academic databases for recent work on educational data mining and learning analytics to ground this paper in current evidence.