Table 1 Experiment result
From: Detecting unregistered users through semi-supervised anomaly detection with similarity datasets
Dataset | Algorithm | Threshold | Train accuracy (%) | Validation accuracy (%) | Test accuracy (%) | F1-Score | AUC (%) |
|---|---|---|---|---|---|---|---|
Digits Mnist | Isolation forest | 0.03 | 79.5 | 75.6 | 78.9 | 78.9 | 79 |
SVM | – | 89.8 | 88.2 | 78.6 | 78.6 | 79.5 | |
MLP-based Auto-Encoder | 4.22 | 96.2 | 89.4 | 92.5 | 92.5 | 92.7 | |
CNN-based Auto-Encoder | 3.79 | 99.7 | 94.8 | 95.2 | 95.2 | 95.2 | |
Auto-Encoder with K-Nearest Neighbor | 63 | 79 | 76.4 | 72.3 | 72.3 | 72.5 | |
CBIR with K-Nearest Neighbor | 95 | 80 | 80.6 | 79.6 | 79.6 | 79.6 | |
ReNet-5 based Multi-Image | 88 | 100 | 95 | 92 | 92.7 | 92.7 | |
Fashion Mnist | Isolation Forest | − 0.02 | 94.4 | 95.8 | 87.2 | 87.2 | 88.03 |
SVM | – | 90.2 | 91.4 | 83.9 | 83.9 | 84.3 | |
MLP-based Auto-Encoder | 4.07 | 95.6 | 92.6 | 87.7 | 87.7 | 87.9 | |
CNN-based Auto-Encoder | 3.77 | 95.9 | 94.8 | 87 | 87 | 87.3 | |
Auto-Encoder with K-Nearest Neighbor | 64 | 81.2 | 78.8 | 78.3 | 78.3 | 78.3 | |
CBIR with K-Nearest Neighbor | 89 | 61.8 | 59.6 | 73.3 | 73.3 | 75.4 | |
ReNet-5 based Multi-Image | 66 | 92.2 | 85.4 | 84.7 | 84.7 | 84.7 | |
Foot Pressure | Isolation Forest | 0.01 | 84.6 | 83 | 51.9 | 51.9 | 53.6 |
SVM | – | 90.1 | 91.4 | 50.2 | 50.2 | 50.62 | |
MLP-based Auto-Encoder | 1.13 | 77 | 73.9 | 58.9 | 58.9 | 59.8 | |
CNN-based Auto-Encoder | 1.09 | 77.6 | 75 | 56.7 | 56.7 | 57.3 | |
Auto-Encoder with K-Nearest Neighbor | 60 | 81.8 | 80.7 | 79.7 | 79.7 | 79.7 | |
CBIR with K-Nearest Neighbor | 69 | 62.6 | 61.3 | 67.4 | 67.4 | 67.7 | |
ReNet-5 based Multi-Image | 78 | 97.7 | 92.7 | 89.2 | 89.2 | 88.95 |