The condition of a vehicle sensor system is an effective indicator used by many other vehicle systems. This article is devoted to the problem of choosing a forecasting method for vehicle sensors. Sensor data are considered as multivariate time series. The aim of the study is to determine the best forecasting model for the type of data under consideration. The LSTM neural network-based method and the VARMA statistical method were chosen for the analysis. These methods are preferred because of their ability to process multivariate series with complex relationships, their flexibility, which allows them to be used for series of varying lengths in a wide variety of scenarios, and the high accuracy of their results in numerous applications. The data and plots of computational experiments are provided, enabling the determination of the preferred option for both single-step and multistep forecasting of multivariate time series, based on the values of error metrics and adaptability to rapid changes in data values.

Keywords: forecasting methods, forecast evaluation, LSTM, VARMA, time series, vehicle sensors system