Researchers are continually exploring ways to enhance diabetes prediction driven by growing global diabetic population and concern within medical community. This has sparked the development of more advanced technologies for early disease detection. Diabetes currently accounts for an average yearly death rate of 38% worldwide. In response this research aims to build a machine learning-based system to improve accuracy of identifying diabetic patients. The authors used Pima Indians Diabetes (PID) dataset to train machine learning models such as random forest, support vector machine, Naïve Bayes, and K-nearest neighbor for predicting diabetes. They evaluated these models using key performance metrics like accuracy, sensitivity, and specificity to determine best-performing algorithm. Random forest model achieved a notable accuracy of 94.21%, significantly surpassing other models in research. Proposed system outperformed current baseline machine learning algorithms used for diabetes prediction on similar datasets.