Modeling PISA 2022 student performance with interpretable fuzzy methods: a comparison of FPM and ANFIS

Abstract

This study explores the potential of the Fuzzy Propositional Model (FPM) for predicting student achievement using process data from the PISA 2022 mathematics tasks in the Turkish sample. The model interprets students’ problem-solving behaviours through rulebased reasoning and triangular membership functions, providing insights into how learning processes unfold rather than focusing solely on correctness. The results indicate that the FPM yields pedagogically meaningful interpretations of behavioural indicators such as response time, number of actions, and task revisits, linking them to varying achievement levels. Although data-driven models like ANFIS may achieve marginally higher numerical precision, the FPM stands out by offering transparent, interpretable rules that enhance educational understanding and support data-informed decision-making. These findings demonstrate that explainable fuzzy logic models can serve as practical tools in large-scale assessments, helping educators and policymakers transform process data into actionable insights about student learning.