The combination of factors including the aging of water distribution infrastructure, growth in water demands, and limited operating budgets have created interest in robust and rigorous methodologies to prioritize rehabilitation and renewal decisions for water distribution infrastructure. The merits of probabilistic network modeling are used to investigate issues of water infrastructure failure and to demonstrate inclusive and dynamic analyses. Decomposable Markov Networks (DMNs) and machine learning techniques in the domain of water distribution systems are developed. A framework is developed that assesses causality or correlation between pipe breaks and relevant factors based on data from the Greater Toronto Area (GTA). The role of factors such cement mortar lining (CML), soil type, pipe material and dimension are employed. The framework can be used to assist decision-making by estimating probabilities of future pipe breakage and identifying rehabilitation options to decrease breakage probabilities. The DMN-based machine-learning approach provides an innovative way to combine engineering knowledge and available data in a robust and formal statistical manner.