Table 1.

Results of statistical comparisons between the two groups (Early PD patients/Controls) for each extracted feature, along with the percentage of times the feature was selected during the leave-one-subject-out validation process.

Feature	NFT		HT		NP
	Statistical Significance	Times selected (%)	Statistical Significance	Times Selected (%)	Statistical Significance	Times Selected (%)
$\overline{{\mu }_i}$	—	—	p < 0.001	100%	p < 0.001	100%
${\sigma }_{{\mu }_i}$	—	—	p < 0.001	3%	p < 0.001	0%
$\overline{{\sigma }_i}$	p < 0.001	0%	p < 0.001	94%	p = 0.013	0%
${\sigma }_{{\sigma }_i}$	p < 0.001	0%	p < 0.001	3%	p = 0.040	0%
$\overline{S_i}$	p < 0.001	91%	p = 0.010	0%	p < 0.001	0%
${\sigma }_{S_i}$	p = 0.136	0%	p < 0.001	0%	p < 0.001	0%
$\overline{K_i}$	p < 0.001	9%	p = 0.028	0%	p < 0.001	0%
${\sigma }_{K_i}$	p < 0.001	0%	p = 0.165	0%	p = 0.247	0%
$\overline{C_U}$	p = 0.248	0%	p < 0.001	0%	p = 0.468	0%
${\sigma }_{C_U}$	p = 0.002	0%	p < 0.001	0%	p < 0.001	0%
$\sum |C_U|$	p < 0.001	0%	p < 0.001	0%	p < 0.001	0%

Statistical significance is computed using the non-parametric two-sided Mann-Whitney U test. The percentage of times each feature is selected is computed over 33 loops of the leave-one-subject-out validation process for the best-performing configuration. Features selected consistently ($>\mathrm{90\%}$ of times) are presented in bold and in all cases, they differ significantly between the two subject groups (p < 0.001). See Supplementary Material Table S.1 and Table S.2 for more detailed information on features.