Daniel Kozelj

Abstract
District Metered Areas (DMAs) are central to leakage control,
but partitions based only on topology or demand ignore local
failure risks. We propose an asset risk weighted spectral
partitioning XGBoost-derived pipe failure probabilities (PFp)
are integrated into a generalized normalized cut (GNC)
framework. The pipe level PFp are length-weighted and
aggregated into nodes to form vertex weights, that guide the
spectral solver to balance clusters by state risk. Using the
Ljubljana-Šentvid case, we compare PFp-weighted GNC with a
demand-balanced baseline across representative edge
weighting cases (unweighted
, diameter  , length  ,
minimum-cost  , and topological case  ), identical
clustering (squared-euclidean). The condition dependent
balancing favors fewer, more compact DMAs, and concentrates
pipes networks poor-condition pipes on smaller DMAs. The
method is reproducible and data-based and embeds PFp directly
into the partitioning to provide hydraulically coherent,
operationally tractable, and risk-oriented DMAs for aging
WDNs.