Machine learning enabled techniques along with electronic health records (EHR) can be used to make predictions about any disease and improve overall health of the patients before it worsens. This research paper provides implementation of different machine learning models for diabetes prediction on an EHR dataset. This was done on a synthetic 30 K Synthia dataset from Harvard Dataverse. We employed models like pre-trained BERT model, logistic regression, support vector machines and random forest algorithms to find out the accuracy by fine-tuning the datasets on it. The findings underscore the necessity of a multi-metric evaluation approach in model assessment and suggest avenues for future enhancements and research.