Influence of the training period on anomaly detection in bridge monitoring. The maintenance of bridge structures while ensuring safety increases the importance of automated monitoring systems. The Regression Based Thermal Response Prediction Method (RBTRPM) from the field of machine learning is a promising approach for the automated detection of structural changes. Structural responses predicted based on temperature measurements using machine learning models are compared with the measured structural behaviour, with deviations identified and evaluated. Although RBTRPM was introduced in 2014, its practical application has been limited. This paper analyses the practical application of RBTRPM using the example of the Itztal bridge. The focus of the study is on the timing of model training and the length of the training period, as both parameters have a significant impact on prediction accuracy. The aim is to implement the method as quickly as possible to enable the early detection of structural anomalies. However, with a short measurement period, the models may not achieve high accuracy. This balance is explored, and corresponding recommendations are provided. It is also noted that extrapolating with such models leads to undesired effects. The study confirms the effectiveness of RBTRPM in practice, though it requires a careful selection of training parameters.