Diabetes mellitus (DM) is characterized by hyperglycemia, a chronic illness caused by inadequate insulin synthesis or an abnormal insulin response. Early identification is crucial since the Centres for Disease Control and Prevention (CDC) anticipate that by 2060, the number of Type 2 Diabetes Mellitus (T2DM) cases among those under 20 would have skyrocketed by 700%. This article offers a thorough approach to diabetes prediction using three datasets: the Pima Indian Diabetes Dataset, the Iraqi Diabetes Dataset, and a medical dataset from Kaggle. Logistic Regression, Decision Tree, Random Forest, SVM, K-Nearest Neighbours, Naive Bayes, Gradient Boosting, and many neural network designs (two-layered neural networks, LSTM, and Bi-LSTM) were among the machine learning models used, along with a voting classifier. The experimental findings demonstrate that machine learning can improve diabetes diagnosis and treatment by demonstrating robust prediction capabilities across models.