Back

Machine learning prediction models for nutrient profiles in food after processing

NARAVANE, T.; Tagkopoulos, I.

2022-09-30 bioinformatics
10.1101/2022.09.28.509827 bioRxiv
Show abstract

The future of personalized health relies on knowledge of dietary composition. The current analytical methods are impractical to scale up, and the computational methods are inadequate. We propose machine learning models to predict the nutritional profiles of cooked foods given the raw food composition and cooking method, for a variety of plant and animal-based foods. Our models (trained on USDAs SR dataset) were on average 31% better than baselines, based on RMSE metric, and particularly good for leafy green vegetables and various cuts of beef. We also identified and remedied a bias in the data caused by representation of composition per 100grams. The scaling methods are based on a process-invariant nutrient, and the scaled data improves prediction performance. Finally, we advocate for an integrated approach of data analysis and modeling when generating future composition data to make the task more efficient, less costly and apply for development of reliable models.

Matching journals

The top 8 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.