Using Lasso-regression to select significant exogenous features in time series forecasting

It is proposed to make the preliminary processing of multidimensional time series and optimize the selection of exogenous variables using the least squares method with regularization, in order to use the obtained results in the SARIMAX model. The analysis of weather and atmospheric data of the Tver Oblast showed that the selection of significant exogenous features using Lasso-regression allows minimizing the forecast error and preventing model overfitting. The results obtained confirm that the correct choice of exogenous variables improves the quality of the predictive model.