Mid-infrared spectroscopy combined with machine learning for rapid detection of caffeine in creatine supplements: Supporting food safety and product integrity.

Creatine (CR) and Caffeine (CAF) are popular and extensively consumed dietary supplements (DS) on a global scale. Although the concomitant use of these substances may augment their effectiveness, it results in the risk of adulteration through mislabeling or the undisclosed addition of CAF. The use of reliable and rapid analytical screening tools is essential to ensure product integrity and protect consumer health. This study explores the application of Attenuated Total Reflectance Fourier-Transform Infrared (ATR-FTIR) spectroscopy combined with Machine Learning (ML) algorithms to detect and quantify CAF in CR supplements. Six commercial creatine brands were analyzed: five were intentionally adulterated with CAF at concentrations ranging from 0.0 % to 20.0 % (w/w), in 2.0 % increments (0.0 %, 2.0 %, 4.0 %, 6.0 %, 8.0 %, 10.0 %, 12.0 %, 14.0 %, 16.0 %, 18.0 %, and 20.0 %), while a sixth, previously unseen brand was used exclusively for external validation. Chemometric models, including Principal Component Analysis (PCA), Support Vector Machine (SVM) and Partial Least Squares (PLS) were performed. Two independent blind tests were conducted to assess the reproducibility and generalizability of the models: the first using three creatine brands included in model training, and the second using the independent sixth brand. The SVM model attained a Sensitivity (SEN) of 100 % and a Specificity (SPEC) of 75 % in the test dataset, whereas it reached 100 % SEN and 100 % SPEC in the blind tests. The PLS model exhibited a moderate quantitative performance, with a Root Mean Squared Error of Prediction (RMSEp) of 3.06 % and a Coefficient of Determination (R2) of 0.75 in the test dataset. In the first blind test, the PLS model showed R2 = 0.7 and RMSE = 2.84 %, while in the second blind test, the PLS metrics were R2 = 0.73 and RMSE = 2.11 %. Overall, these results demonstrate that ATR-FTIR spectroscopy combined with ML provides a reliable framework for screening creatine supplements adulterated with caffeine, supporting its potential use in industrial quality control and regulatory monitoring.