Hybrid Response Surface-Machine Learning Optimization of Doxorubicin-Loaded Chitosan Nanoparticles

Chitosan nanoparticles (CS NPs) are widely explored for drug delivery due to their biocompatibility, biodegradability, and favorable cationic surface properties; however, formulation optimization is often limited by one-factor-at-a-time (OFAT) and trial-and-error approaches that fail to capture complex variable interactions. In this study, a design of experiments (DoE) strategy integrating response surface methodology (RSM) and machine learning (ML) was applied to the development of doxorubicin hydrochloride (DOX HCl)-loaded CS NPs prepared by ionic gelation. Three formulation variables were systematically evaluated for their effects on critical quality attributes (CQAs), including particle size (PS), polydispersity index (PDI), zeta potential (ZP), and encapsulation efficiency (%EE). Multiple ML algorithms, namely, linear regression, k-nearest neighbors, decision trees, bootstrap forests, boosted trees, Gaussian process regression, artificial neural networks, and support vector machines, were comparatively assessed. RSM models demonstrated strong statistical adequacy, while ML analysis identified SVM as the most effective predictive model across all CQAs within the investigated data set. Experimental validation of the optimized formulation revealed close agreement between predicted and observed responses for both modeling approaches. Overall, this hybrid RSM-ML strategy offers a data-efficient and predictive framework for rational NP formulation development, supporting quality-by-design (QbD)-driven pharmaceutical optimization.