Automated Detection of Urban Flooding from News

Although minor overflows do not cause a huge amount of loss; as the number of such overflows increases, the amount of water wasted, and the compound consequent challenges become considerable. Therefore, detecting overflows, investigating their cause root and resolving them in a timely manner are among new needs for infrastructure managers. This paper suggests a new method for detecting distributed water overflows by extracting the flood information (such as the date and location of the incident) from the news. As a case study, we crawled Montreal newspaper and news websites to detect the related news to urban flooding and their detailed information. We trained several classifiers to identify news relevant to flood. Our experiments illustrate that by applying mutual information method for feature selection and employing support vector machine (SVM) architecture as the classifier, relevant news can be detected with an accuracy (F-measure) of above 80%. Such actionable information can help infrastructure managers with a wide range of decisions from repair and maintenance of existing systems, to capacity evaluation for new designs.