The study objective is to develop a system for forecasting the service life of components and parts of railway rolling stock. The task to which the paper is devoted is to determine the service life of wheel pairs of railway rolling stock before their next turning. Research methods include a technique for forecasting the resource of components and parts of railway wheel pairs using a system for estimating expected parameters, including wheel run. The study includes the use of three different machine learning algorithms: linear regression, random forest and gradient boosting. The trained models of each algorithm are shown, as well as the convergence of MSE, MAPE an d R-squared metrics at each iteration of learning. The novelty of the work: the study results make it possible to predict the period of wheel pairs operation with high accuracy and include a feedback mechanism for automating and updating the model, which increases the accuracy of forecasting. Research results: on the basis of the proposed method, a system is developed that allows determining the time interval of operation of rolling stock wheel pairs before their next turning. Conclusions: the proposed method will make it possible to predict the period of operation of certain components and parts under the conditions of a given proving ground until their working condition is restored, based on a system for evaluating expected parameters using machine learning.