Enhancing diabetes risk assessment through Bayesian networks: An in-depth study on the Pima Indian population

Abstract

This study aimed to enhance diabetes mellitus (DM) risk assessment using Bayesian Networks (BNs) and explore their unique capability for risk inference within the Pima Indian population. The primary goal was to move beyond traditional binary DM classification and instead focus on a comprehensive estimation of disease risk, considering the complex interplay of risk factors. The study employed the well-established Pima Indian dataset to assess diabetes risk. BNs were utilized to model the intricate interdependencies among risk factors and provide a nuanced understanding of disease susceptibility, including Bayesian risk reasoning for inferring probabilities of unknown nodes. Logistic regression (LR) was employed as a comparative benchmark to underscore BNs' advantages. BNs demonstrated a distinct advantage over conventional LR, as evidenced by their superior AUC value on the training dataset. This outcome highlighted BNs' capacity to capture intricate variable interactions and perform risk inference, thus enhancing predictive accuracy. The study also showcased BNs' resilience to real-world data distribution nuances, despite a slight decline in AUC on the testing dataset. This research substantiates the potency of BNs in augmenting diabetes risk assessment. The integration of BNs illuminates complex interplay among variables and enables a comprehensive risk evaluation, leveraging Bayesian risk reasoning for nuanced assessments. The study underscores the pivotal role of BNs in elucidating intricate causal relationships, contributing to the broader discourse on leveraging Bayesian methods for health risk prediction. The findings underscore the potential for personalized healthcare interventions aimed at managing DM and mitigating its societal burden.