Design and Evaluation of an Artificial Intelligence–Based Model for Predicting Twelfth-Grade Students’ Mathematics Performance Based on Individual, Educational, and Learning Technology Factors

Abstract

Purpose: This study aimed to design and evaluate an artificial intelligence–based model for predicting the mathematics performance of twelfth-grade experimental sciences students based on individual, educational, and learning technology factors. Methods and Materials: This applied, quantitative, cross-sectional, and predictive study was conducted among twelfth-grade experimental sciences students in Isfahan, Iran. A total of 187 questionnaires were distributed, and after removing incomplete, inconsistent, and invalid records, data from 84 students were retained for analysis. Data were collected using a structured questionnaire assessing individual factors, educational conditions, and learning technology variables, together with students’ recorded mathematics scores. The dataset was divided into training and testing subsets, and repeated five-fold cross-validation was used for internal model development. Multiple linear regression, ridge regression, lasso regression, elastic net regression, decision tree regression, random forest regression, gradient boosting regression, and support vector regression were compared. Model performance was evaluated using mean absolute error, root mean squared error, and the coefficient of determination. Findings: Previous mathematics achievement had the strongest positive correlation with current mathematics performance, followed by mathematics self-efficacy, study regularity, academic motivation, teacher feedback, instructional quality, and educational use of learning technologies. Test anxiety and digital distraction were negatively associated with mathematics performance. Among the evaluated algorithms, gradient boosting regression achieved the best predictive performance, with a test-set mean absolute error of 0.93, root mean squared error of 1.26, and coefficient of determination of .72. Previous mathematics achievement, mathematics self-efficacy, study regularity, teacher feedback, educational use of learning technologies, test anxiety, and instructional quality were the most influential predictors. Adding educational factors increased explained variance from .54 to .65, while including learning technology factors further increased it to .72. Conclusion: The findings indicate that mathematics performance can be predicted with acceptable accuracy by integrating students’ prior achievement, psychological and behavioral characteristics, instructional experiences, and learning technology use within an interpretable artificial intelligence model.