Fuzzy-driven gradient boosting for interpretable cardiovascular disease risk prediction

Abstract

Cardiovascular disease (CVD) continues to be the foremost cause of mortality globally, highlighting the necessity for predictive models that are both precise and clinically interpretable. However, most machine learning (ML) models operate as “black boxes,” constraining their utility for practical medical decision-making. This paper presents a fuzzy-driven gradient boosting (GB) framework that incorporates fuzzy logic into the ML workflow to enhance interpretability while maintaining predictive performance. Employing the fuzzy c-means (FCM) methodology, triangular membership functions (MFs) are developed to convert five principal clinical variables—age, height, weight, systolic blood pressure (ap_hi), and diastolic blood pressure (ap_lo)—into linguistically interpretable categories. Each data point is transformed into a set of fuzzy membership values derived from the pertinent MFs, signifying its partial association with different linguistic categories. The framework is evaluated using 10-fold cross-validation (CV) on three feature types: raw data, fuzzy-linguistic labels, and fuzzy MF values, with the latter being the core fuzzy representation. The experimental results indicate that the hybrid model achieves the best performance, with an accuracy of 73.37% and an area under the receiver operating characteristic curve (AUROC) of 0.800 on Dataset I, and an accuracy of 98.83% with an AUROC of 0.996 on Dataset II. Fuzzy transformation was shown to significantly improve model transparency and provide clinically meaningful explanations. Overall, this proposed framework offers a promising direction for the development of interpretable, reliable, and high-performance artificial intelligence (AI)-based clinical decision support systems (CDSS) for predicting CVD risk.