Table 2.

Variation of number of positive/negative cases in acquired dystonia, and the top features (those that appear in >5 top 10 models) as the threshold on BFMDRS-m to define improvement/no improvement is varied.

Threshold on	Improved	Not Improved	Best features (i.e.,
% change in	(%)	(%)	appeared >5
BFMDRS-m			models out of the
			top 10 models)
0	58 (60.4)	38 (39.6)	CMCT, Baseline Severity
1	58 (60.4)	38 (39.6)	CMCT, Baseline Severity
2	52 (54.2)	44 (45.8)	CMCT, Etiology
3	48 (50.0)	48 (50.0)	CMCT
4	45 (46.9)	51 (53.1)	CMCT, Baseline Severity
5	41 (42.7)	55 (57.3)	CMCT, Sex
>5	Performance degradation (not enough models that perform better
	than majority class classifier)