Development of a classification model for student programming anxiety levels using logistic regression algorithm

Abstract

As programming increasingly becomes a core competency across diverse academic disciplines, mitigating programming anxiety is essential to fostering student engagement, confidence, and academic success in computing-related courses. This study addresses the need for an automated and accurate classification model capable of identifying students currently experiencing programming anxiety. A classification model was developed using the Logistic Regression algorithm, guided by the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework. The methodology involved systematic data preprocessing, feature selection, and the use of the Synthetic Minority Over-sampling Technique (SMOTE) to handle class imbalance. Five supervised classification algorithms were evaluated and compared: Logistic Regression, Support Vector Machine, Naïve Bayes, Random Forest, and Decision Tree. Among these, Logistic Regression produced the best results, achieving an F-measure of 96.77 percent, an accuracy of 97.75 percent, a precision of 96.88 percent, a recall of 96.70 percent, a Cohen’s kappa of 0.950, a mean absolute error of 0.0225, a root mean squared error of 0.1464, a relative absolute error of 0.03 percent, and a root relative squared error of 30.71 percent. The resulting model offers practical value for researchers and educators by enabling the automatic detection of programming anxiety, with strong potential for integration into institutional platforms such as learning management systems and academic support services.