An evolution of statistical pipe failure models for drinking water networks: a targeted review

Abstract The use of statistical models to predict pipe failures has become an important tool for proactive management drinking water networks. This targeted review provides overview the evolution existing models, grouped into three categories: deterministic, probabilistic and machine learning. main advantage deterministic is simplicity relatively minimal data requirements. Deterministic predicting failure rates network or large groups pipes perform well. These are also useful shorter prediction intervals that describe influences seasonality. Probabilistic can accommodate randomness time-to-failure, interarrival times probability failure. Probability individual models. Generally, learning approaches complex more accurately improve predictions yet requires expert knowledge. Non-parametric better suited non-linear relationships between variables. Census socio-economic require further research. Choosing most appropriate model careful consideration type variables, interval, spatial level, response level inference required.