Comparative RSM-ML analysis to predict cellulase production from Lysinibacillus capsici isolate using banana...

A key bottleneck in the sustainable valorization of lignocellulosic biomass remains the efficient, low-cost production of cellulase. Although widely used, Response Surface Methodology (RSM) is limited by its quadratic assumptions, which may not capture nonlinear interactions. Despite progress in using Machine Learning (ML) for cellulase prediction, integrated approaches that combine RSM with interpretable ML methods, along with regression diagnostics, permutation importance, and SHAP-inspired (Shapley Additive Explanations) methods, remain exceedingly scarce in cellulase studies, limiting both predictive performance and biological insight. In this work, a cellulolytic bacterium was isolated from the banana pseudostem and identified as Lysinibacillus capsici through 16S rRNA gene analysis. Box-Behnken Design (BBD) was adopted to optimize cellulase production, considering key process parameters such as pH, temperature, incubation time, substrate concentration, and inoculum volume. To improve predictive capability, a comparative study was conducted on four ML techniques: Artificial Neural Network (ANN), Bayesian Regularized Neural Network (BRNN), Radial Basis Function Neural Network (RBFNN), and Support Vector Machine (SVM). Among the evaluated models, BRNN and RBFNN showed slightly better performance; however, the differences were marginal, indicating comparable overall predictive behavior. Regression plots and statistical metrics validated model robustness, while SHAP-inspired analysis quantified the relative contributions of substrate concentration, pH, and temperature. The findings emphasize that predictive performance is influenced by both the modeling strategy and data availability; therefore, comparisons between RSM and machine learning models should be interpreted in this context. Overall, this integrated RSM-ML framework provides a structured and interpretable approach for optimizing enzyme production from lignocellulosic biomass.