Comparative analysis of automatic time-series forecasting approaches for potato wholesale price index in India

This paper investigates the effectiveness of 11 automatic time-series forecasting techniques in forecasting the wholesale price index (WPI) of potatoes in India. Techniques include autoregressive integrated moving average (ARIMA), error-trend-seasonality (ETS), four artificial neural network (ANN) models, and five hybrid approaches. Evaluation is based on mean absolute percentage error (MAPE). The forecast horizon extends up to 15 months. This work revealed that the ETS-ANN method is the most effective, showcasing an average MAPE of 5.42%. The improvement of the forecast accuracy of the hybrid ETS-ANN over the naive (baseline) is 59.8%, ETS is 29.18%, and ANN is 41.85%. It indicates a significant enhancement in forecast accuracy. The ETS-ANN approach exhibited statistically significant results. It validates the ETS-ANN technique's effectiveness in accurately forecasting the potato WPI in India. It contributes to this specific domain and provides valuable insights for policymakers and stakeholders. Additionally, it may serve as a methodological guide for other agricultural commodities.