Task-Conditioned Synthetic Data Generation for Improving Machine Learning Performance in Agricultural Prediction Tasks
06:00 · July 14, 2026 · arXiv cs.AI RSS
Machine Learning (ML) algorithms have been widely used to estimate agricultural variables across diverse contexts. However, because the quantity and quality of training data strongly influence performance of ML algorithms, their use can be constrained by limited or incomplete reference data. Synthetic Data Generation (SDG) offers a practical approach to address this issue by producing artificial but realistic samples that preserve key characteristics of the original data. Building on teacher-student knowledge transfer and in-context learning for tabular data, this study proposes a Task-Conditioned SDG (TCSDG) algorithm that pairs a Bayesian Network generator with a transformer-based tabular foundation model (TabICL). The proposed algorithm was evaluated on two agricultural prediction tasks: crop yield prediction and crop type classification. Six benchmark SDG algorithms were also utilized to compare their performance with that of TCSDG. Across twelve study sites, two training-data fractions, four multiplication ratios, and three predictive ML algorithms, augmenting the original data with TCSDG-generated synthetic data improved ML performance in 89% of the crop type classification experiments and 74% of the crop yield prediction experiments. TCSDG also substantially outperformed benchmark SDG algorithms and was the only method to consistently improve ML performance across both tasks at the aggregate level. The study demonstrates that carefully designed and processed synthetic data can improve ML performance in precision-agriculture applications. TCSDG offers a practical and extensible framework for generating synthetic data that supports downstream ML agricultural prediction. The full implementation of TCSDG is publicly available as open source at https://github.com/HamidEbrahimy/TCSDG.
Summary
The article presents Task-Conditioned Synthetic Data Generation (TCSDG), a method that addresses the performance limits of supervised machine learning models when agricultural reference data are scarce, incomplete, or unevenly distributed. TCSDG pairs a Bayesian Network generator, which produces a large pool of candidate samples, with TabICL, a transformer-based tabular foundation model that acts as a task-aware teacher. Candidate samples are filtered according to their consistency with the downstream prediction task, then selected through coverage-preserving mechanisms and target-space budget allocation that balance class frequencies without distorting global marginal distributions.
The framework was tested on twelve datasets spanning six crop-yield regression tasks and six crop-type classification tasks, across two training-data fractions, four synthetic multiplication ratios, and three different predictive algorithms. Augmentation with TCSDG data raised performance in 89 percent of the classification experiments and 74 percent of the yield-prediction experiments. In aggregate comparisons it outperformed six established benchmark generators drawn from probabilistic-graphical, adversarial, latent-variable, and diffusion families, and remained the only method that delivered consistent gains across both task types.
The design rests on the premise that useful synthetic tabular samples must be statistically plausible under the observed distribution while preserving the feature–target relationships required by the learner. By embedding teacher-guided filtering and in-context evaluation inside the generation pipeline, TCSDG produces data that directly supports downstream agricultural prediction rather than merely replicating marginal statistics. The full implementation is released as open source.
Why it matters
This research is highly relevant for Dutch AI researchers and AgriTech enterprises, as the Netherlands is a global leader in agricultural innovation. The open-source TCSDG framework provides a rigorous, reproducible method for overcoming data scarcity in precision agriculture, directly applicable to Dutch and EU-wide crop prediction models.
