Nowadays, diabetes is a common disease affecting millions of people worldwide, and it is generally more prevalent among women. Recent health research has adopted various innovative and advanced technologies to diagnose individuals and predict diseases based on clinical data. One such technology is Machine Learning (ML), which enables more accurate diagnosis and prediction. The data used in this study is the Pima Indian women diabetes dataset from Kaggle and the UCI data repository. This study focuses on predicting diabetes using the KNN algorithm model by applying optimization to the dataset using the SMOTE-ENN technique to enhance prediction accuracy for Pima Indian women. The dataset was trained and tested with five different splits using Jupyter Notebook to determine the best accuracy for the KNN algorithm model. Parameters such as classification accuracy, classification error, and the ROC curve were evaluated, along with identifying the variables influencing the risk of diabetes. The results showed that applying SMOTE-ENN optimization to the research dataset significantly improved the prediction accuracy using the KNN algorithm model. With a 70% training and 30% testing data split, the model achieved a classification accuracy of 0.96, a classification error of 0.04, and an AUC of 0.95. These predictions indicated that Pima Indian women are more likely to develop diabetes due to factors such as age above 33 years, the number of pregnancies, excessive sugar consumption, blood pressure, skin thickness, insulin levels, BMI (Body Mass Index), and genetic predisposition to diabetes