Reproduce “Predicting Mycotoxin Contamination in Irish Oats Using Deep and Transfer Learning” on OpenResearch

About this paper

Alan Inglis, Fiona Doohan, Subramani Natarajan, Breige McNulty, Chris Elliott, Anne Nugent, Julie Meneely, Brett Greer and 8 more

Researchers developed and evaluated deep and transfer learning models, including TabPFN, for predicting multi-response mycotoxin contamination in Irish oats, achieving strong classification accuracy and identifying weather patterns and seed moisture as primary predictors. A custom masked loss function was introduced to handle missing mycotoxin measurements effectively.

Key insights

• Pre-trained models like TabPFN can effectively generalize and provide strong predictive performance for complex mycotoxin prediction in agricultural contexts, even with relatively limited domain-specific data.
• The introduction of custom masked loss functions is crucial for enabling the effective utilization of real-world agricultural datasets with prevalent partially observed mycotoxin measurements.
• Weather history patterns (humidity, rainfall, temperature) in the 90-day pre-harvest period and seed moisture content are the most influential environmental and agronomic factors driving mycotoxin contamination in Irish oats.

Summary by alphaXiv.