Non-targeted detection of food adulteration using an ensemble machine-learning model

Area of Science:

• Food Science and Technology
• Analytical Chemistry
• Machine Learning Applications

Background:

• Economically motivated adulteration (EMA) causes severe public health and economic damage.
• Current target-oriented food authentication methods fail to detect novel or unencountered adulterants.
• The risk of widespread contamination, similar to the melamine incident, remains high.

Purpose of the Study:

• To propose and validate an ensemble machine-learning model for non-targeted food adulteration detection.
• To develop a method capable of identifying adulteration without prior knowledge of specific adulterants.
• To assess the model's effectiveness using raw milk as a case study.

Main Methods:

• Development of an ensemble machine-learning model for non-targeted analysis.
• Utilized data from standard industrial testing for raw milk analysis.
• Validated the model's performance through cross-validation with spiked, novel contaminants.

Main Results:

• Achieved high accuracy (0.9924) and F1 score (0.9913) for known adulterant types.
• Demonstrated robust performance with an F1 score of 0.8657 for unencountered adulterants.
• First study to show feasibility of non-targeted detection using standard industrial data.

Conclusions:

• The ensemble machine-learning model effectively detects unprecedented adulteration in a non-targeted manner.
• This approach can identify suspicious samples by uncovering discriminative profiling patterns.
• The technique shows potential for broad application across various food commodities to improve food safety.