The high commercial and nutritional value of saffron has made it a target for widespread adulteration, with safflower being a common substitute due to its physical and sensory similarities. This study presents ATR-FTIR spectroscopy combined with feature selection algorithms and machine learning techniques to detect safflower adulteration in saffron. For the first time, ATR-FTIR is integrated with five feature selection techniques: the Chi-square Test (Chi2), minimum redundancy maximum relevance (mRMR) algorithm, Neighborhood component analysis (NCA), Laplacian Score (LS), and Relief algorithm, to identify influential spectral features for classification. In addition, the number of features, including 0, 250, 500, and 1000 features, was compared. Then, the classification accuracy was evaluated using support vector machine (SVM) and principal component analysis (PCA) models. The highest accuracy was achieved using a combination of data preprocessing methods, including Standard Normal Variate (SNV) and Savitzky-Golay (S-G + D1 + SNV), along with 500 features selected by the mRMR algorithm, yielding 100% accuracy on the training dataset and 98.8% on the testing dataset. The results of this research show the use of FT-IR spectroscopy for accurate and fast detection of safflower fraud in saffron, which is recommended in combination with artificial intelligence to minimize financial incentives for adulteration and increase human food safety and health in the food industry.
