Solubility prediction of edible insect proteins: A comparative study of machine learning and response surface...

This study evaluated whether machine learning models could offer improved predictive performance over traditional response surface methodology (RSM) in predicting the solubility of edible insect proteins according to extraction and processing conditions. The machine learning models included linear regression (LR), decision tree (DT), random forest (RF), and XGBoost (XGB). The RSM model yielded statistically significant outcomes (p < 0.001), its applicability across diverse datasets is limited. Meanwhile, DT, RF, and XGB demonstrated exceptional predictive accuracy and reliability, with R² values greater than 0.99. DT and XGB showed improved metric scores after a 10-fold cross-validation, therefore, compared to RSM, they can be used to predict the unseen data accurately. Additionally, feature importance analysis revealed that the protein extraction methods had a major influence on solubility. These findings suggest that machine learning models can be used to optimize experimental conditions of edible insect proteins research, thus, reducing cost and time.