<p>Affecting millions globally, Parkinson’s disease (PD) is a degenerative neurological disorder that profoundly influences both motor and non-motor functions. Effective treatment depends on early PD detection, which can assist in slowing down the course of the condition and enhance patients’ quality of life. Early-stage PD is difficult to identify, though, because of the variable and complex character of the first symptoms. Designed to use a multimodal dataset, including actual mobility activities, the Uncertainty-Aware Boosting Ensemble (UABE), presented in this work, is a new machine learning framework to improve early PD identification. Using LightGBM models, the UABE framework focuses on uncertain samples, enhancing predictions and increasing resilience and classification accuracy. This paper derived its dataset from the REMAP project. It comprises skeletal positioning data and accelerometer data from PD and non-PD samples obtained in settings similar to home situations. Optuna uses entropy-based uncertainty analysis to handle difficult categorization problems and is applied inside the UABE structure to maximize hyperparameters. The experimental results show that the UABE model performs remarkably with accuracy (100%), 1.0 precision, recall, and F1-Score for both healthy and PD classes. Furthermore, the study of uncertainty reveals that the model exhibits great confidence in most of its predictions, even if few samples are regarded as dubious. With an eye toward a more accurate and quick diagnosis via the UABE framework in clinical settings, this research shows how machine learning might improve early identification of Parkinson’s disease.</p>