Artificial Intelligence-Enhanced Detection of Sesame Oil Adulteration Using FTIR, d13C Analysis, and Symbolic Regression

Abstract

The adulteration of sesame oil poses significant challenges for public health, food safety, and environmental sustainability, requiring advanced analytical and decision-making tools for detection. This study presents an integrated framework combining Fourier Transform Infrared (FTIR) spectroscopy, δ13C isotope ratio mass spectrometry, and artificial intelligence within an environmental engineering perspective. FTIR spectra were processed using an unsupervised deep autoencoder, producing a Spectral Deviation Index (SDI) with excellent discriminatory power (AUC = 1.00). In parallel, a Box-Behnken design quantified δ13C isotopic responses, while machine learning models, including Random Forest (R2 = 0.93) and symbolic regression, accurately predicted adulteration patterns even at low levels. Symbolic regression provided interpretable equations supporting traceability and quantitative decision-making. The hybrid methodology is rapid, non-destructive, and automatable, offering a scalable pathway for sustainable food quality control and environmental monitoring.