This master thesis investigates the use of machine learning for predicting and assessing the risk of railway vehicle component failures. Data used for failure prediction often comes with limitations due to the complex nature of maintenance or sometimes requires investments for the extraction of information. Instead of real-time data, historical data and failure timestamps, easily accessed by organisations, are examined to see if they have the potential to contribute to a more effective maintenance strategy. Datasets used in maintenance often contain censored data and to overcome this problem survival machine learning models were also examined. Therefore both traditional machine learning models and survival machine learning models were evaluated and compared based on their C-index value. The results demonstrate that the survival machine learning models, which incorporate the risk and time-to-event aspects of the data, performed better than the traditional ones regarding the risk ranking of components. Random survival forest had the best result, and a ranking of important features. These findings indicate that there is a potential for survival machine learning, applied to existing historical data used for risk assessment for components failure.