Smart Education: Predictive Analytics of Student Academic Performance Using Machine Learning Models in Weka and Dataiku Systems

Abstract

Smart education and smart universities are based on active use of descriptive, diagnostic, predictive and prescriptive analytics as prescribed by the Gartner’s Data Analytics Ascendancy Model. This paper presents the up-to-date findings and outcomes of the research, design and development project at the InterLabs Research Institute at Bradley University (U.S.A.) aimed at application of a quantitative approach to student academic performance data analytics in general, and innovative Machine Learning (ML) models-based approaches and systems to predictive academic and learning analytics in particular. The goal of this research is to identify the best ML models in the Weka and Dataiku data processing systems based on various forms of student data representation and multiple evaluation criteria for quality of predictive analytics. The analyzed ML models included Support Vector Machine, Naïve Bayes, Random Forest, Random Tree, Linear Regression, Logistic Regression, k-Nearest Neighbors, Multilayer Perceptron, J48, and Decision Stump models. The evaluation criteria for predictive analytics included Correlation Coefficient, Mean Absolute Error, Mean Absolute Percentage Error, Root Mean Squared Error, Root Mean Squared Logarithmic Error, R2 Score for regression ML models and Correctly Classified Instances and Incorrectly Classified Instances for classification ML models. The obtained research outcomes provide a well-validated recommendation about what ML models should be used in student academic performance predictive analytics in smart education and smart universities.