The success of the analysis and design of a Water Network (WN) is strongly dependent on the veracity of the data and a priori knowledge used in the model calibration of the network. This fact motivates this paper in which an off-line approach to verify datasets acquired from WN is proposed. This approach allows separating data of abnormal events from normal-operation without requiring the knowledge of experts limited by the amount of data to be verified. The core of the approach is the combination of a systematic feature selection and an unsupervised classification tool, known as DBSCAN, which does not require setting the number of clusters to be identified. The proposal is applied to datasets acquired from a Mexican water management utility located in the center part of Mexico. The datasets were pre-processed to be synchronized since they were recorded and sent with different sampling times to a web platform. The pressure and flow-rate recorded every ten minutes correspond to dates between 01/07/2019 @ 00:00 and 01/09/2019 @ 23:50. The water network is formed by 90 nodes and 78 pipes and it provides service to approximately 2000 consumers. The data identified as abnormal were validated with the reports of the WDN managers. The abnormal events identified were communication problems, sensor failure, and draining of the network reservoir.