Machine Learning Optimization of Pullulan Production from Banana Peel Extract
Description
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.
Files
Steps to reproduce
a0_BPextract2.mat is the experimental data, which is used to build different hyperparameter-optimized ML models: b1_Bayesian_tree.m, b2_Bayesian_LS_Boost.m, and b3_Bayesian_SVM.m. Finally, the model is selected using c1_model_selection.m. The d1_PDP_plots.m and d4_LIME_Shaply_finalplot.m are for PDP and LIME/SHAP analysis. e1_ANN_multiobj.m is the MOO process. The output for one script is used as input to other, so they need to be executed in sequence.
Institutions
- Karunya University