Table 1.
Summary of studies on behavioral profiling and anomaly detection.
Fig 1.
Sequence to sequence autoencoder architecture.
Fig 2.
Architecture for behavior analytics model.
Fig 3.
Normalized confusion matrix averaged across five stratified folds.
Rows: true class (Normal, Anomalous); columns: predicted Label. Proportions per row (sum to 1). The operating threshold was fixed per fold on validation data.
Fig 4.
Class-conditional reconstruction-error distributions (Normal, Anomalous), averaged over five stratified cross-validation folds.
The vertical line marks the decision threshold.
Fig 5.
Receiver operating characteristic (ROC) curve averaged over five stratified cross-validation folds.
Thin lines indicate individual folds; the bold line shows the mean. Mean AUC = 0.92.
Fig 6.
Model performance metrics.
Table 2.
Concise benchmark against closely related behavior–sequence models.