Table 2 Model performance on anomaly... | Emerald Publishing

Table 2

Model performance on anomaly detection in synthetic access logs (anomaly class metrics shown)

Model	Precision	Recall	F1-score	Accuracy
Deep AE + LightGBM	0.8910	0.2780	0.4238	0.9313
LSTM	0.8851	0.2610	0.4031	0.9476
Rule-Based	1.0000	0.8080	0.8938	0.9981
CPAD	1.0000	0.2400	0.3871	0.9309
Role-Aware Prototype	0.2080	0.2820	0.2394	0.8371
TranAD	1.0000	0.2400	0.3871	0.9925
LogBERT	1.0000	0.2100	0.3471	0.9922
Edge-Aware Transformer GNN (HeteroConv + TransformerConv)	1.0000	0.1900	0.3193	0.9920
Graph Autoencoder (GAE)	0.9367	0.7400	0.8268	0.9969
Variational Graph Autoencoder (VGAE)	0.9737	0.7400	0.8409	0.9972
Meta-Classifier (Proposed)	0.9976	0.8180	0.8989	0.9833

Model	Precision	Recall	F1-score	Accuracy
Deep AE + LightGBM	0.8910	0.2780	0.4238	0.9313
LSTM	0.8851	0.2610	0.4031	0.9476
Rule-Based	1.0000	0.8080	0.8938	0.9981
CPAD	1.0000	0.2400	0.3871	0.9309
Role-Aware Prototype	0.2080	0.2820	0.2394	0.8371
TranAD	1.0000	0.2400	0.3871	0.9925
LogBERT	1.0000	0.2100	0.3471	0.9922
Edge-Aware Transformer GNN (HeteroConv + TransformerConv)	1.0000	0.1900	0.3193	0.9920
Graph Autoencoder (GAE)	0.9367	0.7400	0.8268	0.9969
Variational Graph Autoencoder (VGAE)	0.9737	0.7400	0.8409	0.9972
Meta-Classifier (Proposed)	0.9976	0.8180	0.8989	0.9833

Source(s): Authors own work