Performance comparison of ARIMA and Time Delay neural network for forecasting of potato prices in India

Accurate, timely and adequate forecasting of perishable crops have significant impact on the farmers’ well-being in Indian agriculture. The time series data of these perishable commodities usually violate the assumptions of time-series datasets i.e., linearity and stationarity. In such conditions, the development and selection of the appropriate forecasting models for agricultural commodities plays an imperative role for various policy decisions. In this study, we are focused on comparison of ARIMA (linear) and TDNN (non-linear) models to accurately model the potato price. The inclusion of these nonlinear model in this study handles nonstationary, nonlinear, and non-normal features of datasets simultaneously. The findings revealed that TDNN outperformed ARIMA, and it is regarded as the best fit model in terms of minimal RMSE and MAPE value. The identification of the best forecasting model and accurate forecasting of market prices would help all the stakeholders to take appropriate decisions.