Predicting Student Success: Design and Implementation of a Data Mining Model in WEKA 3.8.5

Abstract

Every academic term, a certain number of students quietly fall behind, and by the time the warning signs are obvious in final grades, it is often too late to help them. Educational institutions collect a great deal of data about their students, from demographics and prior grades to attendance and online activity, yet most of it sits unused. This paper describes the design and implementation of a data mining model built in WEKA 3.8.5 to predict student success early enough for intervention to matter. We assembled a dataset of academic, behavioral, and demographic attributes for a cohort of students and used it to train and compare several well-established classification algorithms, including J48 decision trees, Naïve Bayes, Random Forest, Logistic Regression, and a Multilayer Perceptron. After cleaning the data, handling missing values, and applying attribute selection to identify the most predictive features, we evaluated each model using ten-fold cross-validation. The Random Forest classifier gave the best overall result, reaching an accuracy of 89.3% with an F-measure of 0.89, while the J48 tree, though slightly less accurate at 84.7%, produced transparent rules that educators can actually read and act on. Attribute selection revealed that prior academic performance, attendance, and early-semester assessment scores were the strongest predictors of final outcome, outweighing demographic factors. The study demonstrates that a practical, interpretable early-warning system can be built entirely within WEKA using open tools and modest data, and it offers a workflow that other institutions can adapt. We close with a discussion of interpretability, fairness, and the steps needed to move such a model from analysis into day-to-day academic support.