A significant portion of railway network income is spent on the maintenance and restoration of the railway infrastructure to ensure that the networks continue to provide the expected level of service. The execution of the interventions – that is, when and where to perform maintenance or restoration activities, depends on how the state of the infrastructure assets changes over time. Such information helps ensure that appropriate interventions are selected to reduce the deterioration speed and to maximise the effect of the expenditure on monitoring, maintenance, repair and renewal of the assets. Presently, there is an explosion of effort in the investigation and use of data-driven methods to estimate deterioration curves. However, real-world time history data normally includes measurement of errors and discrepancies that should not be neglected. These errors include missing information, discrepancies in input data and changes in the condition rating scheme. This paper provides solutions for addressing these issues using machine learning algorithms, estimates the deterioration curves for railway supporting structures using Markov models and discusses the results.