Fourier-Transform Infrared Spectroscopy and Chemometric Models for Acrylamide Detection in Cereal-Based Baby Foods: A Machine Learning Approach

Abstract

Monitoring baby cereal foods is vital, as acrylamide levels can pose a food safety risk during infants’ rapid growth phase. In this study, Fourier-transform infrared (FTIR) spectroscopy, covering the spectral range of 650–4000 cm⁻¹, combined with multivariate methods such as interval partial least squares (iPLS), principal component analysis (PCA), and partial least squares discriminant analysis (PLS-DA), was employed to classify and detect acrylamide contamination in the samples at different concentrations (0, 10, and 50 ppm). The models were evaluated using key metrics such as sensitivity, specificity, and classification efficiency. The iPLS regression model, using Savitzky-Golay (SG) filter preprocessing, demonstrated superior performance over the full-spectrum partial least squares regression (PLSR). The root mean square error of cross-validation (RMSECV) for biscuits was 0.106 using the iPLS model and 0.162 for the full-spectrum model, while for Cerelac samples, the values were 0.061 and 0.097, respectively. PCA showed distinct clustering of samples according to acrylamide concentrations. In the PLS-DA model, pure samples showed 100% classification efficiency, while spiked samples (biscuits and Cerelacs) achieved average sensitivity, specificity, and efficiency rates of 95.6%, 100%, and 97.2% and 100%, 93.3%, and 96.5%, respectively. This study shows that FTIR spectroscopy with chemometric models is a reliable, cost-effective, and rapid method for detecting acrylamide in the cereal foods.