Journal: Scientific Reports

Article Title: Proactive soft-failure prediction in optical transport networks via physics-inspired features and Infrastructure-as-Code orchestration

doi: 10.1038/s41598-026-52186-3

Figure Lengend Snippet: System architecture. The multi-physics stochastic simulation and the Ghosh–Adhya real-data benchmark feed a shared feature-extraction pipeline that produces 15-dimensional physics-inspired feature vectors (10 OSNR lags, velocity, acceleration, rolling mean, rolling standard deviation). The Random Forest regressor emits time-to-failure estimates; upon three consecutive sub-threshold predictions (persistence filter), the orchestration layer commits a desired-state change to a Git repository (Fig. ), triggering Kubernetes reconciliation and a Terraform-driven make-before-break migration over NETCONF/OpenROADM.

Article Snippet: The predictor is evaluated on (a) a calibrated multi-physics stochastic simulation spanning five degradation modes and (b) the Ghosh–Adhya (2025) Mendeley optical soft-failure benchmark comprising 756 real lightpaths with OSNR, BER, laser current, and received optical power across 900-sample trajectories for four failure classes.

Techniques: Extraction, Standard Deviation, Migration

Journal: Scientific Reports

Article Title: Proactive soft-failure prediction in optical transport networks via physics-inspired features and Infrastructure-as-Code orchestration

doi: 10.1038/s41598-026-52186-3

Figure Lengend Snippet: Empirical characterization of the Ghosh–Adhya (2025) real-data benchmark (training split, 3,024 trajectories). Percentage of trajectories crossing the 18 dB soft-failure alarm and the 15 dB hard-failure threshold, by class. EDFA and NLI failures produce strong OSNR signatures (52% and 82% hard-threshold crossings respectively); ECL failures are OSNR-invariant due to AGC compensation (0.5% crossings, indistinguishable from no-failure baseline), establishing the scope of an OSNR-based predictor.

Article Snippet: The predictor is evaluated on (a) a calibrated multi-physics stochastic simulation spanning five degradation modes and (b) the Ghosh–Adhya (2025) Mendeley optical soft-failure benchmark comprising 756 real lightpaths with OSNR, BER, laser current, and received optical power across 900-sample trajectories for four failure classes.

Techniques:

Journal: Scientific Reports

Article Title: Deep learning-based phishing classification framework for accurate detection using optimized URL intelligence

doi: 10.1038/s41598-026-46481-2

Figure Lengend Snippet: Overall ADUIN phishing detection framework.

Article Snippet: Dataset , Benchmark phishing URL dataset (PhiUSIIL, Kaggle, Mendeley).

Techniques:

Journal: Scientific Reports

Article Title: Deep learning-based phishing classification framework for accurate detection using optimized URL intelligence

doi: 10.1038/s41598-026-46481-2

Figure Lengend Snippet: Analysis of phishing classification accuracy.

Article Snippet: Dataset , Benchmark phishing URL dataset (PhiUSIIL, Kaggle, Mendeley).

Techniques:

Journal: Scientific Reports

Article Title: Deep learning-based phishing classification framework for accurate detection using optimized URL intelligence

doi: 10.1038/s41598-026-46481-2

Figure Lengend Snippet: Analysis of zero-day phishing detection rate.

Article Snippet: Dataset , Benchmark phishing URL dataset (PhiUSIIL, Kaggle, Mendeley).

Techniques: