Forecasting Agricultural input Price Index Using Statistical and Deep Learning Methods

Abstract

Agricultural Input Price Index is calculated and published by Turkish Statistical Institute each month in order to track the changes in prices of products and services that are used for current agricultural production and future investments. The prediction of the index will enable agricultural producers to make timely decisions regarding investment decisions and product preferences and will increase their competitiveness in the domestic and international markets. In order to predict changes in this index, (ARIMA, SARIMA) and deep learning models (CNN, LSTM) were used in a comparative way in the study. It is known that CNN and LSTM models capture both linear and nonlinear data traits. The prediction results are evaluated by Root Mean Squared Error (RMSE) and Mean Squared Error (MSE) metrics. According to the study results, ARIMA (RMSE: 0.16409, MSE: 0.0269247) and CNN (RMSE: 0.16994, MSE: 0.288791) models achieved the best results, and they are followed by LSTM model.