A Zero-Inflated Bayesian Model for the Prediction of Water Pipe Bursts

The processes and mechanisms giving rise to failures in repairable systems such as underground water pipes are quite complex and not quite fully understood yet. Therefore flexible statistical models that try to explain these processes are necessary. Data available on pipe breaks is usually poor and left truncated resulting in data sets in which a significant number of pipes have no failures recorded at all. Trying to apply flexible point process models such as the NHPP to such data may limit the ability to adequately capture the failure process. In this paper a zero-inflated NHPP is proposed which tries to deal with such problems. The new model is tested on the real-life pipe data set. The results obtained show that although the zero-inflated NHPP did not outperform the NHPP in general, it provided a better fit to the data and slightly more precise results.