Diabetes mellitus (DM), which causes by an imbalance in blood insulin levels, is one of the most prevalent diseases in human history. The DM may cause harm to the immune system, the body, and other internal organs. If appropriate actions are taken at early stages the diabetes can be controlled. Many machine learning (ML) techniques are now employed to classify and predict DM in its early stages. To improve DM classification, a hyper-parameter-based Hard Voting-Based Ensemble Classifier (HVEC) is proposed in this study. Logistic Regression (LR), Extreme Gradient Boosting (XGB), and Random Forest (RF) are the developed HVECs, while Grid Search Algorithm (GSA) is employed in the proposed model for hyperparameter optimization. To facilitate data interpretation by classifiers, the diabetes prediction dataset is preprocessed using standard scalar, label encoding, duplicate record removal, and K-Nearest Neighbor (KNN)-based imputation. Recursive Feature Elimination (RFE) is then used to select and eliminate features which are deemed insignificant and for dimensionality reduction principal component analysis (PCA) is used. Finally, utilizing the reduced features from PCA, HVEC-GSA carries out an effective DM classification. The suggested HVEC-GSA was examined using the F1-score, recall, accuracy, and precision metrics. The HVEC-GSA is compared using ML techniques and different ensemble models. The accuracy of HVEC-GSA for the diabetes prediction dataset is 96.95% which is the higher accuracy in comparison with the existing research.