Alert-Driven Pattern Mining in Action – Reducing Alert Floods in Road Traffic Management

The immense flood of alerts continuously generated in large-scale control systems of critical infrastructures, such as energy or traffic networks, poses a substantial challenge to their efficient and safe operation. Despite ongoing research efforts, mining alert patterns as a means to cope with alert floods remains particularly difficult, since relationships between alerts are often unknown due to the complexity of systems and limitations in log data. In pursuit of a robust solution to this problem, this paper introduces our alert-driven pattern mining approach, which is based on a hybrid, multi-objective evolutionary algorithm. The approach is unique in that it simultaneously maximizes pattern coverage, enabling reasoning about the full range of underlying relationships, and leverages pattern frequency to identify both rare and frequent patterns. This dual focus supports alert flood reduction in regular as well as exceptional, potentially critical situations. The applicability and effectiveness of the proposed approach are demonstrated using real-world log data from the domain of road traffic management, with a particular focus on the characteristics of the discovered patterns.