همایش ملی بیوانفورماتیک ایران

صفحه اصلی / 4th international edition and 13th Iranian Conference on Bioinformatics

A Fuzzy Bayesian Network Model for Personalized Diabetes Risk Prediction: Integrating Lifestyle, Genetic, and Environmental Factors

نویسندگان :

Lida Hooshyar¹ Nadia Tahiri²

1- Graduate researcher (IASBS) 2- Assistant Professor, Department of Computer Science, University of Sherbrooke.

کلمات کلیدی :

Disease risk prediction،Gene-Environment interactions،Fuzzy Bayesian networks،Fuzzy clustering

چکیده :

This study introduces a novel Fuzzy Bayesian Network (FBN) model for predicting diabetes risk by integrating hereditary traits, environmental influences, and personal lifestyle data. Traditional Bayesian networks, which use Boolean logic and set theory, have limitations in handling the complexity and uncertainty of real-world health data. Boolean logic relies on rigid true/false classifications, which oversimplify relationships between variables. Similarly, set theory groups different samples into the same category, failing to capture their differences. These limitations hinder the accurate representation of overlapping and uncertain data. The proposed FBN model overcomes these challenges by assigning unique fuzzy membership degrees to each sample, ensuring distinct representation. The methodology applies the chain rule and Markov Chain Monte Carlo (MCMC) methods to model complex interactions among variables like age, weight, genetic traits, exercise, and sugar intake, addressing both intrinsic and extrinsic diabetes risk factors. Fuzzy membership functions quantify the degree of belonging to fuzzy sets, with these membership degrees acting as weights in the network. By combining these weights with probabilistic values, the model handles uncertainties and improves prediction accuracy. Genetic information is clustered using fuzzy c-means, simplifying the network while preserving essential variability. Compared to traditional binary models, the FBN offers advantages in managing uncertainty and providing probabilistic risk assessments. Boolean models, limited to rigid true/false relationships, fail to capture the complexities of health data, which often exist on a continuum. Results show that FBN delivers more accurate and flexible predictions, making it valuable for early detection, personalized risk assessment, and preventive care strategies. This adaptable model holds potential for application in other chronic disease research and clinical settings.