Day by day, diabetes mellitus has become a typical malady to hu-mankind from youthful to old people. The development of diabetic patients is expanding step by step because of different causes, for example, bacterial or viral disease, harmful or compound substance blend in with the food, immune system response, corpulence, terrible eating regimen, change in ways of life, dietary patterns, natural contamination, and so on. Consequently, diagnosing diabetes is extremely fundamental to spare human life from diabetes. In healthcare services, this scientific procedure is completed utilizing machine learning techniques for decomposing clinical information to build predictive models to do clinical discoveries. This paper presents a diabetes prediction framework for finding diabetes in its beginning phase from patients’ different clinical symptoms. Besides, this paper investigates the significant features in diabetes prediction using Extreme Gradient Boosting (XGBoost) algorithm. It has been shown that the proposed system can predict diabetes more perfectly from patients’ clinical data on different performance metrics (accuracy was 94.23%, sensitivity was 94.74%, and specificity was 93.94%) with feature important scores.