Although many studies have been conducted on machine learning (ML) models for Parkinson's disease (PD) prediction using neuroimaging and movement analyses, studies with large population-based datasets are limited. We aimed to propose PD prediction models using ML algorithms based on the National Health Insurance Service-Health Screening datasets. We selected individuals who participated in national health-screening programs > 5 times between 2002 and 2015. PD was defined based on the ICD-code (G20), and a matched cohort of individuals without PD was selected using a 1:1 random sampling method. Various ML algorithms were applied for PD prediction, and the performance of the prediction models was compared. Neural networks, gradient boosting machines, and random forest algorithms exhibited the best average prediction accuracy (average area under the receiver operating characteristic curve (AUC): 0.779, 0.766, and 0.731, respectively) among the algorithms validated in this study. The overall model performance metrics were higher in men than in women (AUC: 0.742 and 0.729, respectively). The most important factor for predicting PD occurrence was body mass index, followed by total cholesterol, glucose, hemoglobin, and blood pressure levels. Smoking and alcohol consumption (in men) and socioeconomic status, physical activity, and diabetes mellitus (in women) were highly correlated with the occurrence of PD. The proposed health-screening dataset-based PD prediction model using ML algorithms is readily applicable, produces validated results, and could be a useful option for PD prediction models.