Prediction and detection of freezing of gait in Parkinson’s disease from plantar pressure data using long short-term memory neural-networks

Table 5 FOG Detection: One-freezer-held-out cross validation for the 2-layer LSTM model (16 units per LSTM layer)

Participant held out	FOG data	Non-FOG data	Sensitivity (%)	Specificity (%)	Precision (%)	F1 score
1	3454	82,943	83.0	92.5	31.6	0.46
2	9270	110,130	77.2	90.3	40.0	0.53
3	1502	142,157	72.5	92.8	9.6	0.17
6	4240	191,309	85.9	90.2	16.3	0.27
7	33,841	159,787	77.4	89.5	61.0	0.68
8	3640	101,415	86.2	81.2	14.1	0.24
9	683	141,373	92.2	89.8	4.2	0.08
Mean (SD)			82.1 ± 6.2	89.5 ± 3.6	25.3 ± 18.6	0.35 ± 0.20

ISSN: 1743-0003