Data-driven risk assessment on urban pipeline network based on a cluster model.

•This paper proposes an unsupervised risk assessment framework via clustering algorithm which is free of either expert rating as the Indexing models or elementary events probability estimation as the Bayesian Networks methods. It is easy-to-use and scalable for large-scale data.•Instead of treating clustering as preprocessing step to reduce samples as how the previous works have done, this paper builds a novel data mining framework for risk level assignment via clustering with its results validated by a designed statistical test.•A case study is conducted on an urban gas pipeline network consisting of more than 13,000 (over 1700 km) pipelines, demonstrating the capability of the proposed method to recognize and distinguish the patterns representing the different risk level.