MLR and ANN process modeling, optimization, and XAI tools
Description
This dataset includes CCD result data for Echinocandin B, along with codes for developing multiple linear regression and artificial neural network (ANN) models to optimize the production process based on parameters such as pH, dextrose, molasses, and casein. It contains scripts for hyperparameter optimization, genetic algorithm-based ANN model optimization, explainable AI (XAI) for model interpretation, and statistical model comparison. Additionally, the package provides code for generating response surface methodology (RSM) plots, residual plots for model accuracy assessment, and LIME-SHAP plots for model explanation.
Files
Steps to reproduce
a2_Mod_select_FLM: Processes CCD results to generate multiple linear regression models and stores the best-performing model. b1_fitrnet_bayesian: Builds multiple ANN models using CCD data, optimizes hyperparameters, and stores all generated models. b3_fitrnet_bayesian_Opt_Analysis: Selects and stores the best ANN model based on good fitness and low complexity, also generating regression plot. c1_Opt_GA: Applies a genetic algorithm to optimize the final selected MLR and ANN models, generates a report on optimization progress, and stores the optimized process conditions. d1_LIME_Shaply: Utilizes the CCD results and selected ANN model to generate LIME and SHAP explanations for model interpretability. d3_Lime_Shaply_finalplot: Creates a swarm chart using LIME and SHAP explanations. e1_error_comp: Compares the selected MLR and ANN models using CCD data to evaluate statistical differences between the models. e2_Residual_plot: Plots residual deviations for the MLR and ANN models using CCD data to assess model accuracy. f2_RSM_plots_fitlm & f3_RSM_plots_fitrnet: Generate response surface methodology (RSM) plots for the selected MLR and ANN models.
Institutions
- Karunya University