The swift expansion of healthcare data has presented significant hurdles for improving clinical outcomes. Identifying the relationships between patient characteristics, medical records, and disease outcomes using large volumes of Electronic Health Records (EHRs) is one of the challenges faced. Although several machine learning algorithms have been suggested to predict diseases, many existing methods are still not sufficiently accurate. The primary disadvantage of these methods is their emphasis on the prediction algorithm, which often overlooks crucial aspects like feature selection and weighting. Due to these shortcomings, predictive models may not be as effective. To overcome these obstacles, this research seeks to develop an improved framework for predicting diseases by employing machine learning methods and domain knowledge. This paper investigates the utilization of K-Nearest Neighbor's (KNN) algorithm for predicting diseases. It will use domain expertise to help choose and weight relevant features from the large dataset of diabetes data and the resulting performance for accuracy was 72.73%, Precision 62.50%, Recall 55.56%, area under the curve (AUC) 78.31%, and F1-Score 58.82%.................. Keywords: ................Diabetic Disease, Machine Learning, Disease Prediction, feature selection.