TABLE 2.

Results of hyperparameter tuning for the structure of ConvLSTM model.

No.	Width	No. of Conv layers	No. of LSTM layers	Dropout	Sensitivity (%)
					Non-fall	Pre-impact Fall	Fall
1	32	2	2	0.5	88.99	93.31	96.31
2	32	2	2	0.8	91.49	93.31	96.31
3	32	2	4	0.5	91.64	91.21	96.77
4	32	2	4	0.8	92.84	90.79	96.31
5	32	4	2	0.5	92.41	89.12	96.77
6	32	4	2	0.8	90.51	93.72	96.77
7	32	4	4	0.5	94.84	89.54	94.47
8	32	4	4	0.8	91.28	90.38	95.85
9	64	2	2	0.5	90.93	91.63	97.70
10	64	2	2	0.8	91.65	92.89	97.24
11	64	2	4	0.5	88.54	92.05	98.16
12	64	2	4	0.8	85.78	93.51	97.24
13∗	64	4	2	0.5	92.30	93.30	95.86
14	64	4	2	0.8	90.18	91.63	96.77
15	64	4	4	0.5	91.47	89.94	95.85
16	64	4	4	0.8	90.22	89.54	96.31
17	128	2	2	0.5	91.73	93.31	93.55
18	128	2	2	0.8	93.77	88.28	95.85
19	128	2	4	0.5	90.58	92.89	96.31
20	128	2	4	0.8	92.10	90.79	98.16
21	128	4	2	0.5	90.45	94.98	96.31
22	128	4	2	0.8	90.75	91.63	99.08
23	128	4	4	0.5	88.85	94.56	95.85
24	128	4	4	0.8	88.53	89.96	97.24

^∗Finalized structure of ConvLSTM model.