Predicting the quality of cookies with oleogel as a shortening substitute using hyperspectral imaging and artificial intelligence

A total of 36 oleogels, made from various waxes and vegetable oils, were utilized as shortening substitutes in cookies, and machine learning incorporating three boosting models (AdaBoost, GBM, and XGBoost) was applied to predict their effects on the quality attributes of the cookies. Carnauba wax oleogels exhibited the highest melting temperature and contributed to stronger cookie texture, while beeswax oleogels produced cookies with greater spreadability and lower hardness. Similar hyperspectral patterns were observed across different waxes and oils, with distinct clustering based on the level of wax through principal component analysis. By preprocessing the hyperspectral data of oleogels, the performances of three machine learning boosting models were improved to predict cookie spreadability and hardness, achieving an R² value of over 0.95. These findings demonstrate the effectiveness of integrating hyperspectral imaging and machine learning techniques for predicting the quality characteristics of baked goods when various oleogels are used as shortening replacers.