Comparative Evaluation of Exponential Double Smoothing and ARIMA Architecture for Forecasting Area, Yield, and Production of Potato in Odisha, India

Abstract

Potato (Solanum tuberosum L.) is an important food and cash crop in Odisha, India, yet the state fulfils only a small proportion of its domestic requirement through local production. Accurate forecasting of potato area, yield, and production is therefore essential for effective agricultural planning and policy formulation. The present study comparatively evaluated the forecasting performance of Exponential Double Smoothing (EDS), also known as Holt’s Linear Trend Model, and Autoregressive Integrated Moving Average (ARIMA) models using annual time-series data from 1970 to 2023 collected from the Directorate of Economics and Statistics, Odisha, and Five Decades of Odisha Agricultural Statistics. The dataset was divided into training (1970–2017) and testing (2018–2023) periods to assess out-of-sample forecasting accuracy. Model performance was evaluated using Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), diagnostic tests, and Corrected Akaike Information Criterion (AICc). The best-fitted ARIMA models identified were ARIMA (0,2,1) for area, ARIMA (0,1,0) for yield, and ARIMA (1,1,0) for production. However, EDS consistently produced lower forecasting errors than ARIMA for most of the study variables, indicating superior predictive performance. Forecasts generated through EDS for 2024–2026 suggest a gradual decline in potato cultivation area and production in Odisha, while yield is expected to remain relatively stable with slight fluctuations. The findings indicate that EDS is a more reliable and robust forecasting approach for potato statistics in Odisha and may provide useful support for agricultural policy decisions, storage planning, and strategies aimed at achieving regional self-sufficiency in potato production.