SVM-based predictive model for the most frequent structural failure in Bogota sewer system

Abstract

Deterioration models simulate non-inspected sewer pipelines' structural conditions and are used to support strategic asset management. Most of the deterioration models have been constructed based on state ratings (SR) of the infrastructure. However, recent studies have shown that this simplification could provide incomplete information of the network's state, and therefore the SR may not be adequate to develop deterioration models. A support vector machine (SVM)-based modelling procedure was developed to predict the probabilities of structural failures of sewer pipes in urban areas and the reliability of these predictions. We applied this procedure to Bogota's sewer system. The results suggest that classification SVMs are feasible for developing predictive models of structural failures in sewer systems, which can be used to plan the inspections of sewerage networks, giving priority to specific areas where it is most likely to find the failure.