RAPTOR: Resilience-Aware Prediction and Tracking of Operational Risks from Alarm Information Flows

Telecommunication base station monitoring systems trigger alarms in response to network events, degradations, and recoveries. These alarm sequences are crucial for predictive analytics and for quantifying resilience at base stations and across the entire network. Alarms often hide structured information flows that can be revealed by exploring their causality, thereby revealing the dynamism and interdependencies among the infrastructure’s network components. The RAPTOR framework proposed in this work uses temporal clustering, first-order Markov modelling, and resilience metrics to analyse alarm logs and predict alarm behaviour in Radio Access Network (RAN) base stations. Using Dynamic Time Warping (DTW) to identify co-activated alarm clusters and Markov transition matrices to estimate alarm propagation, the proposed framework supports accurate modelling of information flow, next-alarm prediction, and interpretable resilience scores based on alarm entropy, self-loop tendency, and absorption probability. This framework is developed and validated on real-world alarm logs from numerous base stations of a major telecommunication service provider in the United Kingdom. Preliminary evaluation results indicate this approach is a good resource for network managers in infrastructure planning and maintenance scheduling, utilising features such as early warning, causality tracing, root-cause insights, and alarm interpretability.