Adaptive Statistical Decision Support System for Predicting Academic Performance Using Machine Learning

Abstract

Prediction of academic performance became one of the key elements of educational analytics that helps institutions to detect learners who need academic help and support. In this paper, an Adaptive Statistical Decision Support System (ASDSS) for academic performance analytics through machine learning approach is developed and examined. The data set of 500 university students was simulated according to statistically controllable relationship between attendance, assignment scores, quiz scores, midterm examinations and final examinations. For the purpose of statistical analysis and decision-making two indexes – the Academic Performance Index (API) and the Academic Stability Index (ASI) were proposed. The ASDSS framework incorporates descriptive statistical analysis, correlation analysis and machine learning approaches to analyze academic performance. Three classification algorithms: Decision Tree, Random Forest and Logistic Regression were applied and compared in terms of Accuracy, Precision, Recall and F1-score metrics. As a result of experiments, it was shown that the Random Forest classifier provides the best accuracy equal to 95.33%, then comes Decision Tree with 91.33% accuracy and the last Logistic Regression with 82.67%. Thus, the obtained results suggest that ensemble learning algorithms offer the highest predictive ability for academic performance analytics but have also good classification performance. This proposed decision-making framework provides a transparent and statistically sound approach that may help educational institutions in evaluating student progress and making evidence-based decisions. This framework is scalable and may be used for real educational data sets and intelligent academic advising systems.