Abstract:

Abstract This study aims to classify students as successful and unsuccessful regarding mathematical literacy on Programme for International Student Assessment (PISA) 2015 database through data mining methods. The sample consists of all Turkish students who participated in PISA 2015. While data mining methods such as Support Vector Machine, Multi-Layer Perceptron, and J48 were used in data analysis, the data set was evaluated with 10-fold Cross-validation. The evaluation criteria included F-measure, Precision, Recall, Matthews Correlation Coefficient, and Receiver Operating Characteristic (ROC). In the classification of successful and unsuccessful students, analyses were conducted with 13 statistically significant variables according to Chi-SquareAttributedEval, GainRatioAttributeEval, and InfoGainAttributeEval methods. The results showed that the most important variables for classifying successful and unsuccessful students were learning time per week in total, and father’s education level. The highest ROC value was 0.720. When comparing the precision values, the lowest classification value for the Multilayer Perceptron method was 0.645. There was no single method that performed best for all criteria. Researchers should use at least two methods to obtain more accurate results.

Keywords: