Parkinson disease (PD) is a progressive neurodegenerative disease whose diagnosis is not an easy task because of subjective clinical examination and late onset of motor symptoms. Immediate and correct diagnosis is imperative in the prompt intervention and better patient outcomes. This paper introduces a machine learning-based system of earlystage Parkinsonian disorder detection with neuro-motor voice features. There was a publicly available biomedical voice dataset (22 acoustic features, 195 samples) in which each sample was classified as either Parkinson disease or healthy control. Some of the supervised machine learning classifier such as Support Vector Machine (SVM), k-Nearest Neighbors (KNN), Random Forest, Extra Trees and XGBoost were tested. In order to further improve classification performance and minimize falsenegative predictions, a hybrid stacking ensemble model was set forward that uses XGBoost and Extra Trees as base learners and Logistic Regression as a meta-learner. Adequate preprocessing methods like stratified data splitting and feature standardization were used. The results of the experiment show that the proposed hybrid stacking model is better than single classifiers; with an accuracy of 98.31 percent and is able to remove false negative cases of the Parkinson disease. The results suggest the usefulness of ensemble learning in addressing the issue of class imbalance and enhancing diagnostic accuracy, which means voice-based machine learning models might be proposed as a useful decision-supporting instrument to detect Parkinson disease at an early stage. Further research will be conducted on subject-independent validation and multimodal neuro-motor data integration to improve on its generalizability.