Synthetic Data Generation Boosts Agricultural ML Performance.

A new algorithm called Task-Conditioned Synthetic Data Generation (TCSDG) has been developed to enhance machine learning (ML) performance in agricultural prediction. The challenge in agricultural ML often stems from limited or incomplete training data, which directly impacts the accuracy of predictions for variables like crop yield and type. TCSDG addresses this by generating artificial yet realistic data samples that preserve the essential characteristics of the original data. The TCSDG algorithm integrates a Bayesian Network generator with a transformer-based tabular foundation model (TabICL), leveraging teacher-student knowledge transfer and in-context learning. Evaluated across various agricultural prediction tasks, study sites, and ML algorithms, TCSDG consistently improved ML performance. It showed improvements in 89% of crop type classification experiments and 74% of crop yield prediction experiments, significantly outperforming six benchmark synthetic data generation algorithms. This demonstrates the potential of well-designed synthetic data to overcome data scarcity in precision agriculture.