The paper considers the task of improving the transport infrastructure safety through the development of an intelligent simulation model. The focus is on predicting the water levels in lowland rivers to prevent emergencies such as flooding, bridge collapses and other crises that could disrupt transportation networks. The relevance of this research is determined by a variety of factors, including the rise in climatic hazards such as floods and heavy rainfall, industrial threats, and the limited efficacy of traditional approaches to monitoring and forecasting hydrological conditions. The study involves an analysis of actual emergency incidents, which starkly illustrates the urgent need for timely water level predictions. The methodology for developing a decision support system is based on machine learning technologies. The experimental component of the research utilizes data from the Temernik river. A total of 12 machine learning models were evaluated with the objective of selecting the most accurate and efficient models for future forecasting applications. The models were subjected to rigorous statistical assessment using a variety of performance metrics. Following the analysis, the most effective models have been identified and recommended for subsequent deployment and use.