Parkinson’s disease (PD) is a progressive neurological disorder that mostly impacts the speech and the movement of the body of an individual. Early diagnosis can significantly increase the control of symptoms and the quality of life of an individual. The application of machine learning (ML) to voice analysis as an initial diagnosis of Parkinson’s disease is the topic studied in our proposed work. These were speech-related features such as noise to harmonic ratio, jitter, shimmer and pitch variations which were acquired on a dataset of vocal datasets of PD patients and healthy subjects. The most accurate among all the six machine learning models tested was Support Vector Machine (SVM). The cross-validation, feature selection, and preprocessing were used to improve the performance. Based on our results, ML and more specifically SVM is a practical and non-invasive approach to early screening of Parkinson disease.