Abstract
Potato is grown in many states in India like Uttar Pradesh, West Bengal, Bihar, Maharashtra, Karnataka and Gujarat, but its prices are volatile during the post-harvest period. This poses challenges for farmers, consumers and policymakers. Farmers, often needing immediate cash, sell at lower prices, while a lack of timely market intelligence hampers strategic selling and policy decisions, highlighting the need for accurate price forecasting. The selection of the Agra market was done as it has the highest total arrivals of potato among all other markets of India. The weekly price data from January 2003 to December 2023 were utilized for price forecasting, where comparisons were made between traditional statistical models like ARIMA and SARIMA, and neural network models such as ANN and LSTM. It was found that the long short-term memory (LSTM) model outperformed other models, showing the lowest mean absolute percentage error (MAPE) of 7.99% and the highest accuracy. The out-of-sample forecasts for the Agra market using LSTM showed the lowest MAPE and root mean square error (RMSE) values, with forecasted prices for the months of 2024 ranging from Rs 613 to Rs 1702 per quintal. This accurate forecasting of potato prices plays a crucial role in optimizing supply chain management, supporting informed decisions on procurement, storage, and distribution. It enables farmers to sell strategically, helping to mitigate price volatility.
