Diagnostic classification of Parkinson’s disease based on non-motor manifestations and machine learning strategies

Abstract Non-motor manifestations of Parkinson’s disease (PD) appear early and have a significant impact on the quality life patients, but few studies evaluated their predictive potential with machine learning algorithms. We 9 algorithms for discriminating PD patients from controls using wide collection non-motor clinical features two databases: Biocruces (96 subjects) PPMI (687 subjects). In addition, we whether combination both databases could improve individual results. For each database 2 versions different granularity were created feature selection process was performed. observed that most able to detect high accuracy (>80%). Support Vector Machine Multi-Layer Perceptron obtained best performance, an 86.3% 84.7%, respectively. Likewise, led reduction in number variables better performance. Besides, enrichment data moderately benefited performance classification algorithms, especially recall lesser extent accuracy, while precision worsened slightly. The use interpretable rules by RIPPER algorithm showed simply (autonomic olfactory dysfunction), it possible achieve 84.4%. Our study demonstrates analysis parameters through techniques can recall, allows us select discriminative create tools screening.