Optical Network Anomaly Detection Eval

optical-network-anomaly-detection-eval

Machine Learning for Real-Time Anomaly Detection in Optical Networks — Behera et al. (2023) (arXiv:2306.10741, 2023)

What this evaluates

Evaluates the ability of an encoder-decoder LSTM model combined with statistical hypothesis testing to detect unexpected anomalies in optical network quality-of-transmission metrics. It probes whether predicted soft-failure trajectories can distinguish predictable degradation from sudden, anomalous BER deviations in real-time.

Datasets

• Synthetic Optical Network PLM Dataset — total 6081; splits: train (5472), test (609)

Metrics

• Accuracy (primary) — range: [0, 1]
  • Proportion of correctly classified instances (normal vs. anomalous) averaged over 1000 simulations. Calculated as (TP + TN) / (TP + TN + FP + FN).
• Precision — range: [0, 1]
  • Ratio of true positives to all predicted positives: P = t_p / (t_p + f_p).
• Recall — range: [0, 1]
  • Ratio of true positives to all actual positives: R = t_p / (t_p + f_n).
• F-measure — range: [0, 1]
  • Harmonic mean of Precision and Recall: F = 2 * (P * R) / (P + R).

Input / output format

Input: Time-series sequences of Bit Error Rate (BER) values sampled every 1.5 hours, generated via a sliding window of size k=50 with stride s=70 from a simulated 6-node elastic optical network topology.

Output: Binary classification label per sequence/interval (normal vs. anomalous), determined by comparing predicted BER trajectories against actual observations using a scaling parameter theta_tau in a statistical hypothesis test.

Scoring recipe

tp = sum(1 for p, g in zip(predictions, gold) if p == 1 and g == 1)
fp = sum(1 for p, g in zip(predictions, gold) if p == 1 and g == 0)
fn = sum(1 for p, g in zip(predictions, gold) if p == 0 and g == 1)
tn = sum(1 for p, g in zip(predictions, gold) if p == 0 and g == 0)
accuracy = (tp + tn) / (tp + tn + fp + fn)
precision = tp / (tp + fp) if (tp + fp) > 0 else 0
recall = tp / (tp + fn) if (tp + fn) > 0 else 0
f_measure = 2 * (precision * recall) / (precision + recall) if (precision + recall) > 0 else 0
return {'accuracy': accuracy, 'precision': precision, 'recall': recall, 'f_measure': f_measure}

Common pitfalls

• The dataset is fully synthetic and generated via a specific Physical Layer Model (PLM) with Weibull-distributed EDFA degradation, not a public real-world trace.
• Anomaly detection is performed every 5 minutes on sliding windows, but the paper averages accuracy over 1000 independent simulations to report final metrics.
• The scaling parameter theta_tau is tuned empirically to keep false positives near zero, meaning reported accuracy/F-measure are conditional on this specific threshold optimization.

Evidence (verbatim from paper)

Training is performed over a set of 5472 sequences, and testing over the following 609 sequences, demonstrating an accuracy of 1.26\times 10^{-7}. ... In total, 12 anomalies are randomly induced (i.e., a typical percentage of abnormal instances in the dataset) and anomaly detection is performed every 5 minutes ... Specifically, the proposed method achieved the highest accuracy of 92.11% at \theta_{\tau}=1.27 ... To further assess the effectiveness of the proposed anomaly detection model, additional metrics are examined, namely the Precision (P), Recall (R), and F-measure as given below ... P=\frac{t_{p}}{t_{p}+f_{p}};R=\frac{t_{p}}{t_{p}+f_{n}};F=2\times\frac{P*R}{P+R}

Citation

@misc{behera2023opticalanomaly,
  title={Machine Learning for Real-Time Anomaly Detection in Optical Networks},
  author={Behera et al. (2023)},
  year={2023},
  note={arXiv:2306.10741}
}

• arXiv: 2306.10741