Data-Driven Machine Learning Surrogate Models for Predicting Dissolution and Gel Viscosity in Hot-Melt Extruded Polymeric Formulations: A Quality-by-Design Framework

Hot-melt extrusion (HME) is a solvent-free, continuous manufacturing technology that is increasingly central to the production of modified-release and abuse-deterrent oral dosage forms. The relationships among formulation composition, process conditions and critical quality attributes such as drug dissolution and gel viscosity are complex, nonlinear and costly to resolve by experimentation alone. This study develops and benchmarks tree-based machine-learning surrogate models, namely XGBoost, Random Forest and Gradient Boosting, for predicting the 30-min aqueous dissolution of hot-melt extruded polyethylene oxide/hydroxypropyl methylcellulose/carbomer matrices containing pseudoephedrine, and characterizes gel viscosity through exploratory regression. On a small curated dataset (42 records; five features), Random Forest achieved the best held-out performance (R2 = 0.485), with pseudoephedrine drug loading identified as the dominant predictor. Five-fold hyperparameter optimization did not materially improve held-out accuracy, and negative cross-validated R2 values revealed the variance inherent to small samples. Integrating a second data source (96 usable records; 17 features) yielded a markedly stronger optimized model (R2 = 0.984; RMSE = 3.56%), although this gain partly reflects target leakage through the 60-min dissolution endpoint. SHAP analysis attributed predictive power to drug content uniformity and the 60-min readout, and residual diagnostics of the single-source model showed no systematic bias. These results establish a reusable, interpretable, quality-by-design workflow for surrogate modeling in pharmaceutical process engineering and highlight the safeguards needed before such models are used for formulation decisions.

Authors

Publication Details

Journal
International Journal of Science and Research (IJSR)
Published
2026-11-12
DOI
https://doi.org/10.21275/sr261106054406
Primary Topic
Drug Solubulity and Delivery Systems
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Data-Driven Machine Learning Surrogate Models for Predicting Dissolution and Gel Viscosity in Hot-Melt Extruded Polymeric Formulations: A Quality-by-Design Framework

Mohammad Ali Riaz
International Journal of Science and Research (IJSR)
Drug Solubulity and Delivery Systems
article

Data-Driven Machine Learning Surrogate Models for Predicting Dissolution and Gel Viscosity in Hot-Melt Extruded Polymeric Formulations: A Quality-by-Design Framework

Mohammad Ali Riaz
article en

Abstract

Hot-melt extrusion (HME) is a solvent-free, continuous manufacturing technology that is increasingly central to the production of modified-release and abuse-deterrent oral dosage forms. The relationships among formulation composition, process conditions and critical quality attributes such as drug dissolution and gel viscosity are complex, nonlinear and costly to resolve by experimentation alone. This study develops and benchmarks tree-based machine-learning surrogate models, namely XGBoost, Random Forest and Gradient Boosting, for predicting the 30-min aqueous dissolution of hot-melt extruded polyethylene oxide/hydroxypropyl methylcellulose/carbomer matrices containing pseudoephedrine, and characterizes gel viscosity through exploratory regression. On a small curated dataset (42 records; five features), Random Forest achieved the best held-out performance (R2 = 0.485), with pseudoephedrine drug loading identified as the dominant predictor. Five-fold hyperparameter optimization did not materially improve held-out accuracy, and negative cross-validated R2 values revealed the variance inherent to small samples. Integrating a second data source (96 usable records; 17 features) yielded a markedly stronger optimized model (R2 = 0.984; RMSE = 3.56%), although this gain partly reflects target leakage through the 60-min dissolution endpoint. SHAP analysis attributed predictive power to drug content uniformity and the 60-min readout, and residual diagnostics of the single-source model showed no systematic bias. These results establish a reusable, interpretable, quality-by-design workflow for surrogate modeling in pharmaceutical process engineering and highlight the safeguards needed before such models are used for formulation decisions.

International Journal of Science and Research (IJSR)
Drug Solubulity and Delivery Systems
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon