Forecasting tomato production in major Asian producers: a comparative study of ARIMA, exponential smoothing, score-driven models, and XGBoost

Abstract

Abstract Tomato production is a crucial component of the agricultural sector in Asian countries. Accurate forecasting of tomato production is essential for effective agricultural planning, resource allocation, and ensuring food security in the region. This study aims to investigate the patterns and forecast tomato production in five major Asian producing countries: Bangladesh, China, India, Pakistan, and Sri Lanka, utilizing advanced time series models and machine learning techniques. A comprehensive time series dataset spanning from 1961 to 2021 was employed, partitioned into a training period (1961–2014) and a validation period (2015–2021). The study applied various modeling techniques, including ARIMA, Exponential Smoothing, Score-Driven models, and XGBoost. Model performance was evaluated using information criteria, error metrics, and diagnostic tests. Results indicate that while XGBoost yielded the lowest validation errors for several nations due to recent volatility, Exponential Smoothing was selected as the optimal practical model for forecasting Bangladesh’s production to properly account for long-term structural trend extrapolation. Score-Driven models exhibited superior performance for China, India, Pakistan, and Sri Lanka. The selected models generated forecasts up to 2028, revealing continuing upward trajectories for Bangladesh, China, India, and Pakistan, and stabilization for Sri Lanka. This study contributes to the understanding of tomato production dynamics in major Asian producers and offers guidance for agricultural planning, resource allocation, and food security policies. The findings provide valuable insights into the future trends of tomato production in the region, enabling stakeholders to make informed decisions and adapt to potential changes in the agricultural landscape.