Optimizing beta-carotene bioaccessibility predictions with advanced machine learning algorithms and feature selection strategies

Abstract

Since the artificial intelligence is of great attention in food science and technology, in the current study it was aimed to utilize machine learning methods including TreeBoost (TB), Multilayer Neural Network (MLP) and Support Vector Machine (SVM) to predict the bioaccessibility of beta-carotene. The emulsion type (micro- or nano-), oil/water phase ratio, the oil type, the type of the emulsifier (protein or carbohydrate), and beta-carotene concentration were selected as variables for the predicted models. Results demonstrated that TB-based model provide the best prediction for the bioaccessibility of beta-carotene with the values of R2 = 0.4325, RMSE = 17.2484, NMSE = 0.5675, and MAE = 13.3809. Besides, according to TB-based Model 7, the emulsion type (micro- or nano-), oil/water phase ratio, the oil type, the type of the protein, and beta-carotene concentration were found to be efffective on the bioaccessibility of beta-carotene. Based on the comparison of method performance, The RMSE value of TB-based Model 7, showed an improvement of 9.27 % compared to the SVM method and 13.4 % compared to the MLP method. To conclude, the outcomes of this study will shed light to future experiments by simplifying the variables of any process using different machine learning methods.