Skip to main content
. 2021 May 17;21(10):3481. doi: 10.3390/s21103481

Table 5.

The ML prediction of follow-up fall incidents for each sensor location using the TUG tests’ kinematics signals.

Sensor Classification Method Acc (%) Se (%) Sp (%) J Index F1-Score AUC C-Statistic
(95% CI)
C-Statistic
p-Value
Neck SVM_gyro 70.00
(58.19, 72.22)
2.20
(0.00, 20.00)
96.08
(76.92, 100.00)
−0.02 0.16
(0.00, 0.43)
0.50
(0.36, 0.69)
0.34
(0.49, 0.52)
0.367
SVM_accel 69.00
(47.08, 72.22)
1.20
(0.00, 20.00)
95.08
(65.19, 100)
−0.04 0.04
(0.00, 0.26)
0.53
(0.36, 0.68)
2.92
(0.51, 0.54)
0.002
CNN_gyro 60.46
(44.44, 72.16)
42.35
(0.00, 83.5)
67.42
(37.11, 100)
0.07
(−0.37, 0.42)
0.41
(0.00, 0.69)
0.56
(0.33, 0.74)
4.27
(0.54, 0.58)
<0.001
CNN_accel 54.71
(27.78, 72.22)
28.61
(0.00, 100)
64.74
(19.61, 84.61)
−0.06 1 0.26
(0.00, 0.63)
0.46
(0.16, 0.81)
−2.13
(0.43, 0.49)
0.983
Right SVM_gyro 71.78
(66.67, 72.22)
1.40
(0.00, 20.00)
98.77
(88.27, 100)
0.00
(−0.08, 0.12)
0.11
(0.00, 0.31)
0.49
(0.37, 0.60)
−0.25
(0.49, 0.51)
0.599
SVM_accel 70.50
(61.11, 72.22)
0.60
(0.00, 10.50)
97.38
(84.61, 100)
−0.02 1 0.12
(0.00, 0.29)
0.49
(0.31, 0.66)
−0.12
(0.48, 0.51)
0.548
CNN_gyro 50.38
(26.11, 72.22)
44.00
(0.00, 100)
54.10
(0.00, 88.84)
−0.12 1 0.52
(0.00, 1)
0.48
(0.26, 0.68)
−1.83
(0.45, 0.50)
0.966
CNN_accel 49.72
(27.78, 72.22)
43.05
(0.00, 100)
52.11
(0.00, 84.61)
−0.14 1 0.32
(0.00, 0.53)
0.46
(0.18, 0.68)
−2.48
(0.43, 0.49)
0.993
Left SVM_gyro 71.61
(66.67, 72.22)
0.00
(0.00, 0.00)
99.15
(92.31, 100)
−0.01 1 0.00
(0.00, 0.00)
0.49
(0.34, 0.66)
−0.82
(0.48, 0.51)
0.794
SVM_accel 71.06
(61.11, 72.22)
1.00
(0.00, 20.00)
98.00
(84.62, 100.00)
−0.01 1 0.21
(0.00, 0.33)
0.51
(0.37, 0.62)
1.37
(0.49, 0.51)
0.085
CNN_gyro 47.15
(27.78, 77.78)
64.32
(0.00, 100)
40.54
(0.00, 84.61)
0.05
(−0.27, 0.45)
0.54
(0.32, 0.81)
0.41
(0, 0.70)
3.06
(0.51, 0.57)
<0.001
CNN_accel 49.91
(27.78, 72.22)
38.02
(0.00, 100)
54.48
(3.84, 84.61)
−0.08 1 0.28
(0.00, 0.62)
0.44
(0.19, 0.70)
−3.95
(0.41, 0.46)
>0.999

1 Although the range of J is in [−1, 1], there is no practical interpretation of its negative values. Therefore, for the negative mean J index, the confidence interval is not reported.