Parkinson’s disease (PD) is a progressive, chronic neurodegenerative disease which currently affects over approximately six million individuals worldwide, causing patients’ gradual loss of mobility and quality of life. Traditionally, the diagnosis relies mainly on clinical observation and examination. Nevertheless, with the last technological developments, the growing availability of databases, and the development of new artificial intelligence techniques, have opened new opportunities for the early and accurate prediction of Parkinson’s disease. Therefore, in this research, we present a binary classification system using machine learning techniques developed to detect Parkinson's disease based on key biomedical features extracted from voice recordings. These biomarkers were obtained from the “Oxford Parkinson's Disease Detection Dataset” in which we applied feature reduction techniques and evaluated four modern machine learning models. Three of them based on decision trees methods: Random Forest, Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost). Also, we used a recent approach for tabular data: Tabular Attention-Based Probabilistic Few-shot Network (TabPFN). The results showed that the most effective models were CatBoost, and TabPFN both obtaining a F1-Score of 96.6%, being competitive respect the state-of-the-art approaches. These results highlight the potential for improved patient outcomes through earlier intervention and more accurate clinical decision making in the real-world healthcare setting.