Optimal sensor placement (OSP) techniques are important to monitor water distribution networks (WDNs). However, these methods are only feasible under the premise of complete leak location information. In fact, partial leak locations may become lost or may not be recorded. Therefore, in this case, the previous OSP methods are not applicable. To overcome this problem, this paper proposes a new OSP strategy considering that some leak locations are unknown, meaning that they are under semi-supervised conditions. The proposed semi-supervised strategy and the original semi-JMI algorithm are applied to two benchmark networks. The results indicate that the monitoring nodes selected by the semi-JMI criterion will lead to the occurrence of monitoring blind zones, resulting in the degradation of the leak localization performance. The addition of the fuzzy c-means (FCM) clustering method will compensate for this problem and improve the accuracy and the stability of leak localization. (C) 2019 Elsevier Ltd. All rights reserved.