Big Data Analytics for Cross-Domain Anomaly Detection to Identify Hidden Service-Impacting Patterns in Fiber Access Broadband Networks

Fibre-to-the-x networks are observed through several operational systems, yet service-impacting degradation often remains hidden because each system describes only a fragment of the end-to-end service. Optical power, line errors, dynamic bandwidth allocation, IP-session behaviour, customer-premises telemetry, topology, alarms and complaints may each remain within local thresholds while their joint movement signals a developing fault. This review examines how cross-domain data correlation and anomaly detection can expose such weak, distributed evidence before it becomes a widespread outage. It synthesises research published from 2020 to 2025 on multivariate time-series modelling, graph learning, feature-domain interaction, optical-network failure management and explainable detection. The review proposes an evidence-centred architecture that aligns heterogeneous telemetry by service, topology and time; learns both within-domain signatures and between-domain dependencies; models propagation across shared network resources; and converts anomaly scores into service-impact hypotheses that engineers can verify. Particular attention is given to incomplete labels, changing baselines, class imbalance, encrypted traffic, topology reconfiguration and the distinction between statistical abnormality and operational harm. The analysis argues that effective FTTx assurance requires neither a single universal model nor indiscriminate data fusion. It requires a layered design in which deterministic rules, unsupervised detectors, graph-temporal models and human validation cooperate under explicit data-quality and governance controls. The resulting framework supports earlier detection, sharper fault-domain isolation and more defensible prioritisation while preserving operational explainability.