Using explainable artificial intelligence and machine learning to predict student academic performance

The aim of this study is to predict students' academic performance using machine learning methods and to make the developed models interpretable through explainable artificial intelligence (XAI) techniques. Today, large amounts of data are generated in educational institutions through student information systems and digital learning environments. The analysis of these data using machine learning methods enables early prediction of students' academic performance and the identification of at-risk students. Explainable artificial intelligence approaches support making model outcomes more transparent, interpretable, and reliable by revealing which variables these predictions are based on. Within the scope of the study, a dataset was created using multidimensional variables that are considered to affect students' academic performance. The data used in the study consists of student information system (UBYS) records and survey data obtained from students. The dataset includes various variables such as students' exam scores, attendance, use of online learning resources, social media usage habits, technology use, and socio-demographic characteristics. During the data preprocessing phase, missing data were handled, categorical variables were transformed into numerical form, and the dataset was prepared for modeling. In the study, both classification and regression-based machine learning models were used to predict students' academic performance. In classification analyses, the Support Vector Machine (SVM) algorithm demonstrated the highest performance. The model achieved an accuracy of 0.9071, an F1-score of 0.9129, a balanced accuracy of 0.9062, an ROC-AUC of 0.9389, and a PR-AUC of 0.9294 on the test dataset. These results indicate that the model can classify students' academic performance with high accuracy and effectively distinguish between classes. In regression analyses, students' academic performance was treated as a continuous variable, and different regression models were compared. The results showed that the XGBoost regression model achieved the best performance. The model obtained an R² value of 0.6963, an RMSE of 12.4614, and an MAE of 8.9904 on the test dataset. These findings indicate that the model can explain a substantial portion of the variance in students' academic performance and produce predictions close to the actual values. One of the key contributions of this study is the interpretation of the developed machine learning models using explainable artificial intelligence methods. For this purpose, SHAP and LIME techniques were employed. According to the SHAP analysis results, the most important variables affecting students' academic performance include father's education level, course (study) notes, use of online courses, the impact of social media usage on studying, and the use of e-learning resources. The results of the LIME analysis indicate that variables such as regular attendance, father's education level, frequency of breaks during studying, participation in online courses, and the place of residence play a significant role in individual predictions. The findings reveal that students' academic performance is associated not only with academic indicators but also with multidimensional variables such as socio-demographic factors, learning behaviors, and the use of digital learning resources. Moreover, explainable artificial intelligence methods have made the decision-making processes of machine learning models more understandable and demonstrated their potential to support data-driven decision-making processes in education. In conclusion, this study shows that machine learning and explainable artificial intelligence approaches provide effective tools for predicting students' academic performance and can contribute to the development of early warning systems in educational institutions.