Performance of time-series forecasting and machine learning techniques to forecast price of agricultural products

This paper studies forecasts of agricultural product prices using Trend Analysis method, Decomposition method, Double Exponential Smoothing method, HoltWinter's method, and ANNs method. Collected data from 2005-2021 to forecast and compare the accuracy between the forecast values and actual values with MAPE, which forecasts for the years 2015, 2015-2016, 2020 and 2020-2021 as a guide for making decisions about choosing the appropriate data for forecasting the price of agricultural products in the future accurately. Experimental results reveal that The ANNs methodforecast more accurately in cases where the basic time-series methods were not good at forecasting, and the results obtained were satisfactory. There is no guideline to tell that in which situation a particular time series method will be more accurate than others. Anexperiment on various forecasting methods and adjustment of related parameters must be conducted to obtain the accurate forecast. In real data sets, there may be some unusual events that cause an inaccurate forecast.