Application of machine learning techniques in time series analysis of prices of pulses

Abstract

Time series analysis of prices helps to capture the movement, trend and seasonality in price series which is useful to different stakeholders, such as farmers, consumers and policy makers. In order to model the structure of prices of pulses, monthly Wholesale Price Indices (WPI) from January 2005 to March 2019 consisting of 171 time series observations were collected from office of the Economic Adviser, Ministry of Commerce & Industry, Govt. of India. The WPI captures the changes in the price level at the initial stages of transaction and government periodically changes the base year to improve the representativeness. To make the WPI series comparable, linking factor were calculated using the average ratio of overlapping monthly prices. Using R software, time series decomposition was carried out to estimate trend and seasonal components in the price series. Seasonal indices revealed that price indices of pulses were on higher side during July to December months. Further, time series models were built to capture and predict the price indices of pulses.