Genetic programming:  a novel method for the quantitative analysis of pyrolysis mass spectral data.

Area of Science:

• Chemometrics
• Machine Learning
• Analytical Chemistry

Background:

• Multivariate data analysis is crucial in chemometrics.
• Pyrolysis mass spectrometry (PyMS) generates complex datasets.
• Accurate quantification of food adulteration is an ongoing challenge.

Purpose of the Study:

• To introduce a novel genetic programming (GP) technique for multivariate data analysis.
• To apply GP for the quantitative analysis of orange juice adulteration using PyMS data.
• To compare GP performance against established methods like partial least squares (PLS) and artificial neural networks (ANNs).

Main Methods:

• Dimensionality reduction of input space by ranking variables based on correlation or mutual information.
• Utilizing genetic programming (GP) for predictive modeling.
• Analysis of orange juice adulteration data with varying percentages of sucrose solution.

Main Results:

• GP achieved predictive errors comparable to or better than PLS and ANNs.
• The GP method facilitated interpretation of input-output variable correlations.
• Reliable quantification of orange juice adulteration (0-20% sucrose) with an RMS error of approximately 1% was demonstrated.

Conclusions:

• Genetic programming provides an effective and interpretable method for analyzing PyMS data.
• This GP approach enables accurate quantification of orange juice adulteration.
• GP offers a competitive alternative to existing multivariate analysis techniques in food analysis.