Towards Accurate Crop Yield Forecasting with Quantum Machine Learning Models

Addressing the persistent issue of food insecurity—particularly in regions experiencing agricultural deficits—requires accurate and timely forecasting of crop yields. Reliable yield predictions are vital for policymakers in key agricultural regions, as they facilitate strategic planning for the redistribution of surplus commodities through international trade. This, in turn, contributes to regional food security and enhances the economic stability of exporting nations.

Among the most widely cultivated crops globally are rice, wheat, maize, soybean, and pigeon pea. Yield prediction for these crops has traditionally relied on various statistical and machine learning approaches. Recently, Quantum Machine Learning (QML) methods—particularly Quantum Neural Networks (QNNs)—have been proposed as a novel paradigm for forecasting applications due to their potential to capture complex, high-dimensional relationships in large datasets. QNNs, leveraging qubits and quantum gates, offer computational advantages over classical models in specific contexts, especially in handling non-linear interactions and entangled feature spaces.

This study aims to forecast the yields of the aforementioned five major crops for the years 2025, 2026, and 2027 using historical data spanning from 1961 onward. The dataset, curated from the Food and Agriculture Organization (FAO) of the United Nations, comprises officially reported statistics from national bureaus across the five leading producing countries for each crop. A training-to-testing split of 75:25 was utilized to develop and evaluate predictive models.

Several baseline models—including conventional statistical regressors and classical machine learning algorithms—were implemented and benchmarked against quantum counterparts, namely Quantum Neural Networks (QNNs) and the Variational Quantum Regressor (VQR). Model evaluation was conducted using standard performance metrics, including Root Mean Squared Error (RMSE), Relative Root Mean Squared Error (RRMSE), and Minkowski distances (with p-values of 1, 2, and 3).

The results demonstrate that quantum models, particularly QNNs, exhibit competitive or superior performance in yield forecasting tasks compared to classical models, highlighting their potential as effective tools for data-driven agricultural decision-making in the quantum era.