Modeling PISA 2022 Student Performance with Interpretable Fuzzy Methods: A Comparison of FPM and ANFIS

Abstract In this study, we aim to predict achievement using the Fuzzy Propositional Model (FPM) with operation-based student data for the mathematical literacy multiplicity group questions in the PISA 2022 Turkey sample. FPM predicts achievement by interpreting student processing behaviours using triangular membership functions and rule bases. Within the scope of the study, observations of a total of 919 students with complete transaction data were analysed.The findings show that FPM produces lower error rates compared to the classical regression model, and provides pedagogically meaningful and interpretable results on student profiles. It has also been observed that the prediction results of FPM provide more process-based and behaviour-based explanations compared to the official achievement scoring of PISA.On the other hand, the data-driven ANFIS model run on the same data set showed a better prediction performance in terms of RMSE value (0.666) compared to the FPM. Bootstrap analysis with 1000 iterations of RMSE comparison showed that the difference between the two models was statistically significant (95% CI: [0.297, 0.306]). This reveals that ANFIS has higher prediction accuracy. However, FPM can still be considered as a worthy option, especially in educational settings, due to its explainability and lower resource requirements.