Diabetes disease is one of the main healthcare challenges in all the world. Undiagnosed diabetes can increase the danger of cardiac stroke, diabetic nephropathy, and other disorders. Early detection of diabetes is necessary to take care of a healthy life. Nowadays, social media is a new dimension to deal with health care by exploiting the real-time shared patients' data to early detect diabetes disease. Furthermore, technologies typically associated with digitalization add value in healthcare, including artificial intelligence, data analytic technologies, and stream processing technologies. Therefore, in this research, we propose a real-time system for predicting diabetes disease from health-based social streaming data to indicate the current status for patient health. The proposed system aims to find the most accurate machine learning model which has the highest accuracy of diabetes prediction. We have used three types of feature selection techniques to select the most relevant features from the used dataset i.e., Recursive Feature Elimination, Univariate feature selection, and Feature Importance. Also, we have evaluated and compared four machine learning models with selected and full features i.e, , Random Forest, Support Vector Machine, Decision Tree, and Logistic Regression Classifier. The experimental results have determined that the random forest model has achieved the greatest accuracy among other models at 84.11%. For online prediction through social media, we have performed our proposed system to handle streaming Twitter data about patients' health. In doing so, Kafka and Spark streaming are integrated into the backend of the proposed system. Then, the random forest classifier is used to predict the patient's current health status in real-time.