Analysis of time series characteristics using machine learning model and correlation matrix in the tasks of forecasting the state of forest ecosystems

Abstract

This paper considers the problem of choosing the required set of characteristics for time series forecasting. The method of solving this problem on the basis of correlation matrix is proposed. A correlation matrix is constructed based on the prepared data, after which a list is formed for each variable, ordered by decreasing modulus of the correlation degree. Then linear regression models are trained and the quality of predictions for different sets of variables from the sorted list is compared. Next, a comparison is made for different values of sampling time to determine the optimal value for each variable. To apply the considered algorithm, information from various measuring sensors taking readings of climate variables in the forest lands was used.