Abstract The paper investigates feed-forward deep neural networks (DNNs) for estimating dynamic resilience of water resource system affected by unpredictable and dangerous events. Besides different architecture of DNNs, hyperaffect the performance of DNNs. The aim of this research was to investigate the capabilities of DNNs in domain of water resources resilience estimation to provide significantly better results than currently developed ANN models from literature. The DNN models were trained and tested using large, generated dataset related to the Pirot water system. In order to generate data, an appropriate model of system dynamics was used alongside MonteCarlo simulations. The dataset contained two hazardous events: flood and earthquake defined in wide range of situations (nearby 2,000), from moderate to severe ones. The efficacy of examined DNNs were evaluated using average error metric as well as time required for training and execution. Deep neural network models for dynamic resilience estimation of a complex water system under hazards info:eu-repo/semantics/conferenceProceedings https://unilib.phaidrabg.rs/o:5126 ISBN: 978-86-81037-79-9 application/pdf 2596598 bytes University of Kragujevac, Serbia Ćirović, Vukašin Stojković, Milan Milivojević, Vladimir 2024 3rd Serbian International Conference on Applied Artificial Intelligence (SICAAI), Book of Abstracts All rights reserved eng Keywords: deep neural networks, dynamic resilience, water resources