This study aims to develop a method for predicting the size of deviations beyond standard values in railway tracks. Despite significant progress made in the predictive maintenance of extended objects, effective adaptation of time series forecasting methods to the estimation of deviation sizes remains limited. The proposed method involves a statistical analysis of multidimensional feature space, formation of input, output, and control parameters, creation a uniform two-dimensional grid, and predicting the size of deviations in the grid nodes. The last step incorporates a linear regression model with L1 regularization, which proved to be a practical approach for prediction deviations, avoiding overfitting, and enhancing the model's generalization ability by adding a complexity penalty. The resulting model can predict the size of deviations based on a significant number of features, allowing for early detection of potential issues and timely preventive maintenance.