Predicting heat-induced hazardous compounds in roasted almonds via browning index and moisture using machine learning

While roasting is essential for developing the characteristic flavor of almonds, it can also result in the formation of thermal contaminants that cause potential threats to human health. This study investigated the effects of roasting conditions (110-150 °C, 10-30 min) on precursor degradation, antioxidant activity, and formation of acrylamide (AA) and 5-hydroxymethylfurfural (5-HMF) in almonds. Higher roasting intensity increased browning index (BI) and total phenolic content, while reducing sugars and asparagine declined. AA and 5-HMF contents peaked at 150 °C/30 min. A support vector regression (SVR) model using L*, a*, b*, BI, and moisture content accurately predicted the levels of AA and 5-HMF (R² > 0.94, RMSE < 10%). Moderate roasting conditions (≤ 130 °C for ≤15 min) are recommended to achieve an optimal balance between product safety and quality. These results suggested that machine learning is an effective method for food risk assessment.