Multifaceted Understanding of Student Success: A Data Mining Approach

This research explores the potential of Data Mining in Education, particularly in the capability to predict student performance and identify the factors that influence academic outcomes. Employing the UC Irvine Student Performance dataset, the study investigates the efficiency of data mining strategies for predicting student outcomes and the co-relating elements. Amongst the nine ML models evaluated, including an ANN model, the linear SVM emerged as the optimal choice, achieving a significantly higher accuracy of 94.9% and an ROC AUC score of 0.959 with a Logarithmic loss of 1.74 on the test set compared to the other candidates. Beyond prediction, the study examines the influence of various factors on academic performance using the coefficients of the Logistic Regression model. Apart from the two mid-term grades, history of failures, internet access, external school support and the area of residence influence the student’s performance in the final examination. By interpreting these findings, the research lays the groundwork for developing data-driven and personalized learning strategies, potentially encouraging a more fair and effective learning environment for all students.