Background: Type 2 Diabetes Mellitus (T2DM) is an increasing global health challenge, requiring strong predictive models for prompt intervention. This study sought to create and validate a logistic regression-based predictive framework for diabetes risk utilizing electronic health record (EHR) data. A retrospective cohort of 10,000 adults devoid of previous diabetes was derived from anonymized electronic health records (EHRs). Demographics, vital signs, laboratory biomarkers, comorbidities, and medication history were all possible predictors. Data preprocessing included dealing with outliers, filling in missing values, and making features more consistent. We used logistic regression with elastic net regularization and divided the data into training, validation, and independent test sets. We used AUROC, AUPRC, calibration, Brier score, and decision curve analysis to figure out how well the model worked. The model got an AUROC of 0.81 and an AUPRC of 0.46 on the test set. It also had good calibration and subgroup consistency. Logistic regression was easier to understand than machine learning comparisons, but it still had similar levels of accuracy. An understandable, EHR-based logistic regression model offers a useful and clinically significant method for predicting diabetes risk. Future research should broaden validation efforts across diverse populations and investigate the integration of advanced AI methodologies.