Improved FTIR-based classification for food authentication using a topological ensemble framework.

Area of Science:

• Food science and technology
• Analytical chemistry
• Data science

Background:

• Food authentication faces challenges with subtle compositional differences.
• Fourier-transform infrared (FTIR) spectroscopy is promising but complicated by spectral artifacts.
• Traditional chemometric methods struggle with nonlinear spectral data.

Purpose of the Study:

• To develop a robust framework for food authentication using FTIR spectroscopy.
• To address limitations of traditional methods in handling complex spectral data.
• To improve classification accuracy and interpretability in food analysis.

Main Methods:

• Integration of sliding-window embeddings with topological data analysis (TDA).
• Application of bagging-based ensemble learning with transparent base learners.
• Evaluation on six benchmark FTIR food datasets from the UCR Time Series Classification Archive.

Main Results:

• The proposed framework demonstrated consistently robust performance across diverse food categories.
• Achieved high classification accuracy and Macro-F1 scores compared to 20 other classifiers.
• Multi-scale topological descriptors enhanced interpretability and provided insights into spectral variability.

Conclusions:

• The framework offers a balance of transparency, robustness, and computational efficiency.
• Provides a reproducible benchmark for future studies on FTIR-based food authentication.
• Successfully enhances the reliability of FTIR spectroscopy for food quality and authenticity assessment.