Automatic classification of municipal call data to support quantitative risk analysis of urban drainage systems

• Published in: Structure and infrastructure engineering 2013-02, Vol.9 (2), p.141-150
• Main Authors: ten Veldhuis, J.A.E., Harder, R.C., Loog, M.
• Format: Article
• Language: English
• Subjects: automatic classification; urban drainage; urban flood risk
• ISSN: 1573-2479; 1744-8980
• DOI: 10.1080/15732479.2010.535543

Quantitative analyses of urban flood risks are often limited by lack of data on flood incidents. Call data are a valuable source of information about urban flood incidents, yet the unstructured nature of call information results in large time investments to prepare the data for application in quantitative analyses. Consequently, the existing call databases are not used for this purpose. If automatic classification routines can be applied to transfer unstructured call data into a quantitative data source, large stores of currently unused data can be made available for quantitative risk analysis of urban infrastructure systems. This article aims to assess whether automatic classification of calls from municipal call centres can reach sufficient accuracy to allow for use of the results in quantitative risk analysis. This is illustrated by the application of automatic classification results in quantitative fault tree analysis for urban flooding, for two cases with datasets of approximately 6000 calls. The results show that the obtained classification accuracy is sufficient to correctly rank failure mechanisms according to their contributions to the overall failure probability. This is a promising first result that shows the potential of automatic call classification to obtain data about failure incidents that are otherwise hard to find.