Abstract
As the second largest potato producer globally, reliable forecasts of output for India and major growing states are crucial. This study developed autoregressive integrated moving average (ARIMA) models alongside state space and gradient boosting machine learning techniques for annual potato production spanning 1967–2020. Model adequacy was evaluated using information criteria, errors metrics and out-of-sample validation. The chosen models provide the following forecasts: India is predicted to produce around 46,712 thousand metric tons, Uttar Pradesh 13,900 thousand metric tons, West Bengal 11,544 thousand metric tons, Bihar 7710 thousand metric tons, Madhya Pradesh 3478 thousand metric tons, Gujarat 3621 thousand metric tons and Punjab 2870 thousand metric tons over the period 2021–2027. While no consistent superior approach emerged, tailoring models to capture data complexity and patterns for each state proved essential for generalization. Quantitatively assessing linearity, stationarity and outliers during model specification is key for stakeholders and policymakers needing precise predictions.