In Odisha, potato is a key crop that experiences significant price fluctuations due to variations in production and market arrivals. This study focuses on the Banki Market in the Cuttack district, a critical trading hub, to forecast potato prices and arrivals, aiding farmers in crop planning. Data from April 2019 to May 2024 on potato arrivals and prices were analysed using machine learning models for prediction estimation. After outlier removal, artificial neural network (ANN) models, specially the time delay neural network (TDNN) model, and support vector regression (SVR) models with varying hidden layer nodes were fitted. Diagnostic test, the Box–Pierce tests were performed to validate the assumptions of error independence and normality. Models that met these criteria were evaluated using root mean square error (RMSE) and mean absolute percentage error (MAPE) to identify the best fit. TDNN(5,8) was chosen as the best model for arrival series with an improvement of 4.3% RMSE and 1.9% MAPE error accuracy in the testing set, whereas SVR(Lags = 10, C = 200, ε = 0.01, γ = 0.001) was chosen as the best model for price series with an improvement of 37.13% RMSE and 68.75% MAPE. The forecast values of both arrival and price of potato are found to be approximately increasing with time with ups and downs at different intervals. We are of the opinion that this study will provide the necessary literature to inform the decision-making process regarding the sale or cold storage of potatoes, as well as the anticipated increase in profits for farmers, thereby achieving economic development.
Full publication URL
