DEEP LEARNING TECHNIQUES FOR FORECASTING ONION MARKET PRICES

Author (s) : A.KOWSALYA, G. SREEDEVI

---

Abstract :
Forecasting agricultural commodity prices is critical for farmers, governments, and agribusiness industries to make informed business decisions by managing risk and uncertainty. One of the most difficult aspects of onion price forecasting is its nature of high price volatility, production uncertainties, export policy, trader hoarding, poor market intelligence, inadequate storage and processing facilities, lack of robust demand and supply, and so on. By improving the accuracy of a suitable onion price forecasting model, policymakers and/or government bodies can make informed decisions based on market demand and supply. Although statistical time series models are highly interpretable, they struggle to achieve high precision in the complex agriculture scenario. In recent years, there has been a surge of interest in machine and deep learning techniques, which have gained prominence as a result of numerous applications in the fields of time series modeling and forecasting, as well as agriculture. There is a scarcity of research in the agricultural domain that is solely focused on onion price forecasting using a deep learning framework. This study contributes to addressing the problem of long-term historical reliance on onion market prices by analyzing and forecasting weekly prices of major consumer markets in India, namely, 30 Delhi, Mumbai, and Bangalore, using recent advancements in deep learning models such as long short-term memory (LSTM) and its variants, attention mechanism based LSTM, convolution neural network (CNN), hybrid CNN-LSTM, temporal convolution neural network, and temporal convolution neural network (TCN).Furthermore, they are also compared with one traditional statistical time series model viz., generalized autoregressive conditional heteroscedastic (GARCH) and two classic machine learning algorithms viz., time delay neural networks (TDNNs) and support vector regression (SVR). The findings show that deep learning models outperform statistical and machine learning models in terms of forecast accuracy measure criteria for all market, and they successfully address the onion price forecasting issue.

Keywords : onion price, deep learning.

Conference Date : 13/11/2020

Pages : 343

Paper ID : 20343