Performance Evaluation of ARIMA and TDNN Models in Forecasting of Potato Prices in Punjab

The present study evaluates the forecasting performance of the Autoregressive Integrated Moving Average (ARIMA) model and the Multilayer Artificial Neural Network, specifically the Time Delay Neural Network (TDNN) model, for predicting potato prices in Punjab, India. Monthly potato price data were sourced from the AGMARKNET portal, covering five major markets within the state, namely Jalandhar, Amritsar, Khanna, Patiala, and Dharamkot. The forecasting performance of the ARIMA and TDNN models was assessed using the Root Mean Square Error (RMSE) as the evaluation metric. The optimal ARIMA model for each price series was identified based on the lowest RMSE observed for the testing set, which revealed that the TDNN model outperformed the ARIMA model. Both the ARIMA and TDNN models identified the Khanna market as the most accurate for price forecasting among the six markets. However, it was noted that the forecasting accuracy of the TDNN model diminished as the forecast horizon extended from 12 to 36 months.