Type 2 diabetes, traditionally considered a condition of older adults, is increasingly prevalent among young adults, posing a significant health risk if undiagnosed. Conventional machine learning models often exhibit biases towards majority age groups, leading to underdiagnosis in younger populations. This study introduces a novel framework to enhance fairness and accuracy in predicting type 2 diabetes among young adults. We propose three advanced algorithms: Age-Personalized Ensemble Learning (APEL), Age-Weighted Neural Network (AWNN), and Synthetic Minority Age Sampling combined with a Hybrid Classifier (SMASH). Unlike traditional models, our approach trains separate predictive models for each 5-year age band (30–34, 35–39, 40–44), ensuring age-specific learning and reducing digital ageism. Using clinical and lifestyle features—including age, gender, BMI, waist circumference, fasting glucose, HbA1c, blood pressure, cholesterol levels, activity level, smoking status, and family history—our methodology demonstrates improved diagnostic accuracy for underrepresented age groups. Experimental results highlight substantial reductions in prediction bias and enhanced model reliability across age bands. This research provides a practical and ethical approach for age-sensitive diabetes diagnosis, supporting clinicians in early detection and personalized healthcare delivery for young adults.

Keywords: Type 2 diabetes, young adults, digital ageism, machine learning bias, age-personalized models, ensemble learning, neural networks, minority sampling, hybrid classifier, healthcare analytics.