Change point-driven interrupted time series and machine learning models for forecasting indian food grain production

Summary

This study develops interrupted time series and machine learning models to forecast Indian food grain production, incorporating a 1987 structural change point identified via Pettitt test. Interrupted artificial neural networks substantially outperformed conventional ARIMA approaches, capturing nonlinearities in production data and achieving error reductions of up to 99% in training. Despite a negative intervention effect in 1987, the overall trend showed sustained production growth across all major grain categories, with insights applicable to policy-making in developing economies facing agricultural system transitions.

Regional applicability

This study is specific to India's agricultural production system and may have limited direct applicability to United Kingdom conditions, which operate under different climatic, institutional, and policy frameworks. However, the methodological approach—combining change-point detection with interrupted machine learning models—could inform UK agricultural forecasting, particularly if structural shifts in policy or practice (e.g., post-Brexit trade changes or climate adaptation) warrant similar time series decomposition.

Key measures

Production forecasts (lakh tons) for rice, wheat, coarse cereals, total cereals, pulses, and total food grains; model accuracy metrics; error reduction percentages; Pettitt test p-values; intervention effect coefficients

Outcomes reported

The study developed interrupted time series and machine learning models to forecast Indian food grain production, identifying 1987 as a significant structural change point. Interrupted artificial neural network models achieved superior accuracy in forecasting rice, wheat, pulses, cereals, and total food grains, with error reductions up to 99.06% in training datasets.

Topic tags