ARCHIVES

Being able to predict a student’s academic performance prior to its occurrence enables a school to prevent it, for instance, by providing extra tutoring and personalized lessons to improve the weak areas. Early identification of at-risk students is becoming a critical success factor for academic advising, yet institutions are often ill-prepared to understand which predictive modeling technique works best for their data. This study compares the predictive accuracy of six different supervised machine learning algorithms including Decision Tree, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Naïve Bayes and Multi-Layer Perceptron (ANN). The study aims to define which machine learning algorithm can effectively predict whether a student will pass or fail a class by analyzing their demographical, social and academic data and which attributes contribute the most to a student’s academic performance. In order to assess the accuracy of predictive models, a standard student dataset was used and 10-fold cross validation performed. Each algorithm was evaluated based on the percentage of correctly predicted outcomes, precision, recall, F1-score and training time. The study revealed that a Random Forest algorithm achieved the highest prediction accuracy (91.2%) followed by an ANN (89.5%), SVM (87.6%) while Naïve Bayes showed the lowest accuracy (75.4%). The number of previously obtained poor grades, failed classes and weekly studying hours appeared to be the most informative features in terms of student performance prediction for all algorithms. Overall, the study demonstrated that ensemble tree-based algorithms seem to be the most effective when it comes to building an early warning system for student performance, albeit with higher computational complexity and institutions should choose which technique to base their prediction model on depending on their specific circumstances.