Skip to main content
. 2022 Sep 19;22(18):7094. doi: 10.3390/s22187094

Table 2.

Comparing a new anomaly detection algorithm with prior methods. The algorithm’s performance in gait detection (values in parentheses indicate the STD between the different validation folds).

Accuracy (%) Precision (%) Sensitivity (%) Specificity (%)
HYAs
FB 71.0 36.3 77.8 69.5
AR-iHMM 73.2 40.0 94.6 68.4
DCNN 96.0 90.9 83.8 98.4
Bidir-LSTM 90.9 70.6 84.3 92.3
New Anomaly Detection 99.4 79.1 44.6 99.4
OAs
FB 76.2 (2.4) 15.4 (5.8) 59.8 (6.3) 77.3 (2.5)
AR-iHMM 85.2 (8.1) 27.3 (8.6) 78.7 (17.9) 85 (9.43)
DCNN 96.2 (4.2) 61.4 (3.4) 69.8 (9.8) 97.6 (4.9)
Bidir-LSTM 81.7 (6.6) 38.8 (11.9) 81.7 (5.7) 81.8 (7.4)
New Anomaly Detection 76.2 (0.5) 29.9 (0.5) 67.6 (0.1) 78.1 (0.6)
PD
FB 77.6 (5.7) 15.2 (11.3) 57.4 (15.6) 57.4 (5.5)
AR-iHMM 92.6 (4.7) 38.1 (24.0) 50 (23.3) 95.1 (5.0)
DCNN 91.7 (7.4) 40.9 (6.0) 72.9 (7.4) 92.9 (9.2)
Bidir-LSTM 76.6 (4.6) 38.3 (7.1) 88.6 (10.9) 76.5 (6.1)
New Anomaly Detection 83.3 (3.8) 20.9 (2.3) 74.7 (8.2) 83.8 (5.4)