Parkinson's disease (PD) is an age-related neurodegenerative disorder affecting millions of elderly people world-wide. The early and accurate diagnosis of PD with available treatment might delay neurodegeneration and prevent disabilities. The existing diagnosis method such as brain scan is an expensive process. The use of speech recognition with machine learning technologies for the diagnosis of PD patients could be less expensive. In this work, we have worked with the voice recorded dataset from UCI machine learning repository. Several studies were performed to identify PD patients from the healthy individuals by using voice recorded data with machine learning algorithms. In this paper, we have proposed an optimized approach of data pre-processing that enhances prediction accuracy for diagnosing PD. We obtain 97.4% prediction accuracy with higher sensitivity, specificity, precision, F1 score and kappa value by using AdaBoost. These improved performance evaluation metrics indicate, the use of voice recording with our optimised machine learning approach is highly reliable in prediction of PD. This approach may have significant implications for early stage diagnosis of PD in a cost-effective manner.