Aflatoxin M1 in milk products: QuEChERS–HPLC-FLD quantification, occurrence patterns across product categories and interpretable risk-based screening

Aflatoxin M1 (AFM1) in milk is a significant concern for food safety. This study (i) measured the occurrence of AFM1 across four milk products and (ii) developed interpretable artificial intelligence / machine-learning tools to support risk-based compliance screening and the prioritization of confirmatory testing. Samples were screened using enzyme-linked immunosorbent assay (ELISA) and confirmed using the quick, easy, cheap, effective, rugged, and safe (QuEChERS) extraction method in conjunction with high-performance liquid chromatography–fluorescence detection (HPLC-FLD). The validated method exhibited a high level of linearity (R² > 0.999), excellent recoveries (80.1–87.5%), low relative standard deviation (RSD% ≤ 9.1) and low limits of detection (LOD) and quantification (LOQ) of 5 and 16.6 ng/L, respectively. In total, AFM1 was confirmed in 52.3% of samples, quantified in 36.8%, and exceeded the 50 ng/L Jordanian/EU regulatory limit in 8.79%, with a maximum concentration of 144 ng/L. Exceedances were concentrated predominantly in fresh milk, with occasional occurrences in milk powder. Ridge logistic regression models delivered excellent detection performance (receiver operating characteristic area under the curve (ROC-AUC = 0.917)) and a moderate performance in predicting rare non-compliance occurrences (ROC-AUC = 0.743; average precision = 0.163). This generated a risk equation and map to optimize dairy surveillance.