Machine learning models to predict micronutrient profile in food after processing.

Area of Science:

• Food Science
• Nutritional Science
• Computational Science

Background:

• Accurate nutritional information for cooked foods is vital for manufacturers and consumers.
• Current methods for determining nutrient profiles face challenges in precision, scalability, and generalizability.
• Existing solutions like analytical methods, retention-factor based methods, and kinetic models have significant limitations.

Purpose of the Study:

• To develop an alternative solution for predicting the micronutrient profile of cooked foods from raw food composition.
• To create a versatile prediction model applicable to multiple food types and cooking processes.
• To demonstrate the efficacy of machine learning in improving nutrient profile prediction accuracy.

Main Methods:

• A machine learning model was trained using an existing food composition dataset.
• Data scaling and transformation techniques were applied to mitigate yield bias.
• The model's predictions were compared against the baseline retention-factor method.

Main Results:

• The developed prediction model achieved a 31% lower average error compared to the retention-factor method across various foods, processes, and nutrients.
• The study highlighted the importance of data preprocessing, including scaling and transformation, for model performance.
• Machine learning demonstrated superior potential over existing methods for nutrient profile prediction.

Conclusions:

• Machine learning offers a more accurate and scalable approach to predicting cooked food micronutrient profiles.
• The findings underscore the significance of data quality and standardization in food composition databases.
• This study provides valuable insights for future food composition data generation and utilization.