Figure 12
The two side-by-side grouped bar charts are labeled “(a)” and “(b)”. In both charts, the horizontal axis lists the metrics “Accuracy”, “Precision”, “Recall”, and “F 1”, and the vertical axis is labeled “Score” ranging from 0.65 to 1.00 in increments of 0.05. Each metric group contains six bars corresponding to the models “Li Conv Former”, “T C N”, “C N N”, “Bi L S T M”, “G C N”, and “T S G C N”. In panel “(a)” titled “Video Dataset”, the bars show approximate values where Li Conv Former achieves about 0.75 accuracy, 0.77 precision, 0.75 recall, and 0.755 F 1; T C N shows around 0.80 accuracy, 0.81 precision, 0.80 recall, and 0.798 F 1; C N N records about 0.75 accuracy, 0.763 precision, 0.75 recall, and 0.75 F 1; Bi L S T M shows about 0.81 accuracy, 0.817 precision, 0.808 recall, and 0.81 F 1; G C N reaches approximately 0.85 accuracy, 0.86 precision, 0.85 recall, and 0.85 F 1; and T S G C N shows the highest values around 0.875 accuracy, 0.88 precision, 0.875 recall, and 0.875 F 1. In panel “(b)” titled “Vibration Dataset”, the bars indicate Li Conv Former with about 0.775 accuracy, 0.77 precision, 0.775 recall, and 0.77 F 1; T C N with approximately 0.825 accuracy, 0.853 precision, 0.825 recall, and 0.817 F 1; C N N with about 0.82 accuracy, 0.818 precision, 0.82 recall, and 0.818 F 1; Bi L S T M with around 0.925 accuracy, 0.927 precision, 0.925 recall, and 0.925 F 1; G C N with roughly 0.70 accuracy, 0.70 precision, 0.70 recall, and 0.70 F 1; and T S G C N with the highest scores near 0.95 across accuracy, precision, recall, and F 1. Note: All numerical data values are approximated.Performance comparison of anomaly detection models under single-sensor conditions. (a) Video data performance comparison. (b) Vibration data performance comparison. Source(s): Figure created by authors