A Novel Metric to Evaluate the Association Rules for Identification of Functional Dependencies in Complex Technical Infrastructures

Functional dependencies in complex technical infrastructures can cause unexpected cascades of failures, with major consequences on availability. For this reason, they must be identified and managed. In recent works, the authors have proposed to use association rule mining for identifying functional dependencies in complex technical infrastructures from alarm data. For this, it is important to have adequate metrics for assessing the effectiveness of the association rules identifying the functional dependencies. This work demonstrates the limitations of traditional metrics, such as lift, interestingness, cosine and laplace, and proposes a novel metric to measure the level of dependency among groups of alarms. The proposed metric is compared to the traditional metrics with reference to a synthetic case study and, then, applied to a large-scale database of alarms collected from the complex technical infrastructure of CERN (European Organization for Nuclear Research). The results confirm the effectiveness of the proposed metric of evaluation of association rules in identifying functional dependencies.