𝗖𝗿𝗼𝗽𝗚𝗔𝗡: 𝗔 𝗖𝗼𝗻𝗱𝗶𝘁𝗶𝗼𝗻𝗮𝗹 𝗚𝗔𝗡 𝗙𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 𝗳𝗼𝗿 𝗦𝘆𝗻𝘁𝗵𝗲𝘁𝗶𝗰 𝗧𝗮𝗯𝘂𝗹𝗮𝗿 𝗗𝗮𝘁𝗮 𝗔𝘂𝗴𝗺𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 𝗶𝗻 𝗖𝗿𝗼𝗽 𝗥𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗦𝘆𝘀𝘁𝗲𝗺𝘀

Can artificial intelligence help farmers make better crop choices when agricultural data are limited, imbalanced, or biased?

This study introduces CropGAN, a conditional Generative Adversarial Network framework designed to generate synthetic tabular data for crop recommendation systems.

Crop recommendation models rely on soil and climatic variables such as nitrogen, phosphorus, potassium, soil pH, humidity, temperature, rainfall, soil texture, and crop type. However, limited, imbalanced, and regionally biased datasets can reduce the reliability of these models, especially in data-scarce farming communities.

To address this challenge, the authors re-engineered the traditional GAN architecture for agricultural tabular data. CropGAN combines class conditioning, minority-aware sampling, and Wasserstein loss with gradient penalty to generate realistic and diverse multi-crop recommendation samples.

The framework was trained on a dataset of 5,000 samples covering 10 crop classes and evaluated against SMOTE and Variational Autoencoder methods using statistical data-quality metrics and classification-performance results.

Although SMOTE better preserved the original data distribution, CropGAN introduced greater diversity and improved downstream model performance. Models trained on CropGAN-generated data achieved the strongest classification results, with the Support Vector Machine reaching 99.4% accuracy.

This research contributes to precision agriculture, synthetic tabular data generation, crop recommendation systems, class-imbalance handling, adversarial learning, and AI-driven decision support for data-scarce farming regions.

📖 Read the full article here:
https://doi.org/10.46481/jnsps.2026.3291

Published in: Journal of the Nigerian Society of Physical Sciences