A Fuzzy Logic-based Approach for Improving the Accuracy and Stability of Machine Learning Models in Educational Data Analysis

Abstract

The increasing adoption of machine learning techniques in educational analytics has created opportunities to predict student performance and support data-driven decision-making. Nevertheless, conventional machine learning models often struggle to manage uncertain, incomplete, and imprecise information that commonly exists in educational datasets. Fuzzy Logic provides an efficient solution by representing uncertain information through linguistic variables and rule-based reasoning, thereby improving model interpretability and robustness. This research proposes a Fuzzy Logic-based framework for enhancing the accuracy and stability of machine learning models used for educational data analysis. Student information related to daily screen time and academic performance was collected and preprocessed before being evaluated using a Mamdani Fuzzy Inference System. Input variables were transformed into fuzzy membership functions representing linguistic categories such as Low, Medium, and High. A knowledge-based rule set was then applied to infer academic performance, and the resulting fuzzy output was converted into a crisp prediction using the centroid defuzzification method. Experimental analysis indicates that the proposed fuzzy model effectively captures the nonlinear relationship between screen time and academic achievement while handling uncertainty more efficiently than conventional statistical techniques. The rule-based reasoning process also enhances model transparency, making the prediction process easier to interpret by educators and researchers. The findings demonstrate that integrating fuzzy logic with machine learning improves predictive reliability, supports informed educational decision-making, and provides a flexible framework for analysing complex real-world datasets.