Exploring lipid profiles as predictors of diabetes risk through machine learning techniques

Diabetes mellitus (DM) is one of the global health challenge characterized by multifactorial pathogenesis and significant morbidity. Dyslipidemia, particularly the Triglyceride to HDL (TG:HDL) ratio, is increasingly recognized as a potential marker for diabetes risk. This study investigates the predictive potential of lipid profiles, focusing on the TG:HDL ratio, using advanced Machine learning techniques. A cross-sectional dataset of 1,000 participants was analysed, incorporating Logistic regression, Decision trees, Random forests, Support vector machines, and Gradient boosting models. Among these, Gradient boosting demonstrated the highest accuracy (88%), with the TG:HDL ratio emerging as a key predictor across all models. Exploratory data analysis revealed a strong correlation between elevated TG:HDL ratios and diabetes prevalence, supported by robust model performance metrics such as precision, recall, and AUC-ROC scores. The findings highlight the clinical utility of TG:HDL ratio as an accessible, cost-effective biomarker for early Diabetes detection and signifies the role of Machine learning in advancing personalized healthcare strategies.