Parkinson’s disease (PD) is a progressive neurodegenerative condition that impairs motor function; hence, prompt identification is essential for successful management. This work looks into the use of speech features from the publicly accessible Parkinson’s Disease Speech Signal Features dataset on Kaggle to identify Parkinson’s disease using supervised and semi-supervised learning models. An ensemble of XGBoost and Support Vector Machine (SVM) classifiers were used in the supervised technique, which produced an accuracy of 90% with 91% precision, 82% recall and 85% F1-score. With an F1-score of 94.91% and a validation accuracy of 94.95%, the semi-supervised model that included Generative Adversarial Networks (GANs) greatly outperformed the other models. The findings show that whereas supervised techniques yield dependable detection, the semi-supervised method provides better accuracy and generalisation, particularly in situations with a limited amount of labelled data. This work contributes to the advancement of Parkinson’s disease detection using speech data by highlighting the possibility of semi-supervised learning in medical diagnosis.