Fig. 3

Effect of sampling rate. Model performance (AUROC) for (a) Binary and (b) Multiclass models of tremor and bradykinesia, using the previously determined sensor set for each symptom. Shaded regions depict a 95% confidence interval on the average AUROC centered at the original sampling rate. Decreasing sampling rate reduces ability to classify tremor beyond 20–30 Hz, with only slight impact on classifying bradykinesia. A 30-Hz sampling rate is sufficient to classify both symptoms using the BioStampRC sensor (Sensor) or smart watch (Watch)