A new approach to the diagnosis of Parkinson's disease (PD) is demonstrated, based on the analysis of UV-induced skin autofluorescence using machine learning methods. The study included 123 participants (23 controls, 16 patients with nonspecific neurological disorders and 84 patients with PD). Spectral data was processed by the principal component method (reduction from 150 to 30 features) and classified by five algorithms. The best results were achieved by a fully connected neural network (ROC-AUC = 0.80, sensitivity =0.92) and logistic regression (ROC-AUC = 0.79). The results support the potential of non-invasive skin spectroscopy as a screening tool for the early diagnosis of PD in clinical practice. Key words: Parkinson's disease, fluorescence, optical spectroscopy, skin, aberrant proteins, advanced glycation end products, machine learning, neural networks, diagnostics. Keywords: Parkinson's disease, fluorescence, optical spectroscopy, skin, aberrant proteins, advanced glycation end products, machine learning, neural networks, diagnostics.