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

Table 9 Percentage of time frozen in one-freezer-held-out cross validation with the 2-layer LSTM model for FOG detection

Participant held out	True positive	True negative	False positive	False negative	Model time frozen (%)	True time frozen (%)
1	2868	76,729	6214	586	10.5	4.0
2	7154	99,394	10,736	2116	15.0	7.8
3	1089	131,869	10,288	413	7.9	1.0
6	3642	172,641	18,668	598	11.4	2.2
7	26,201	143,062	16,725	7640	22.2	17.5
8	3137	82,302	19,113	503	21.2	3.5
9	630	126,898	14,475	53	10.6	0.5
Mean ± SD					14.1 ± 5.2	5.2 ± 5.5

ISSN: 1743-0003