Statistical Analysis of Chemical Element Compositions in Food Science: Problems and Possibilities

Area of Science:

• Food chemistry
• Analytical chemistry
• Statistical analysis

Background:

• Food data analysis often overlooks compositional pitfalls, leading to inaccurate results.
• Classical statistical methods can yield arbitrary outcomes when applied to compositional datasets.
• Accurate chemical analysis is crucial for food authentication, particularly for susceptible products like honey and saffron.

Purpose of the Study:

• To compare compositional data analysis (CoDa) with classical statistical analysis for food chemical component investigation.
• To demonstrate the impact of pre-processing, missing value imputation, and non-detect replacement on analytical outcomes.
• To highlight the differences in results obtained from compositional versus non-compositional methods.

Main Methods:

• Application of compositional data analysis (CoDa) techniques.
• Comparison with classical statistical analysis methods.
• Evaluation of pre-processing strategies, including missing value and non-detect replacement.
• Assessment of Principle Component Analysis (PCA) and Artificial Neural Networks (ANNs) performance.

Main Results:

• Log-ratio analysis within CoDa provided better separation between pure and adulterated food samples.
• CoDa methods enhanced the interpretability and classification accuracy of food authentication data.
• Artificial Neural Networks (ANNs) performed poorly without proper CoDa pre-processing.
• Pre-processing steps significantly influenced PCA and classification outcomes.

Conclusions:

• Compositional data analysis (CoDa) is recommended for analyzing food chemical elements and authenticating food products.
• Log-ratio analysis offers superior performance for distinguishing genuine from adulterated food items.
• Proper pre-processing within CoDa frameworks is essential for accurate classification and analysis.