Machine Learning Optimization of Pullulan Production from Banana Peel Extract

Batch experimental data comprise two numeric (inoculum size and BPE sugar equivalent) and one binary (pretreatment category) predictor, as well as two responses, i.e., EPS and biomass concentration. The data is augmented with Gaussian error addition. The augmented data is used to develop machine learning models, including ANN, RF, support vector regression (SVR), and least-square boosting (LS Boost), which were employed to predict biomass and pullulan production. MOO of the process response is adopted. In addition, barplots of the experimental data, model selection with different degrees of augmentation, regression plots, and XAI model plots are included.