Machine Learning-Based Prediction of Tablet Dissolution Profiles

Abstract

Tablet dissolution testing is a critical quality attribute in pharmaceutical manufacturing because it directly influences drug bioavailability and therapeutic performance. Conventional dissolution studies require repeated laboratory experiments, making formulation optimization time-consuming and resource-intensive. This study identifies a research gap in the limited use of hybrid machine learning frameworks that simultaneously integrate formulation composition and manufacturing process variables for dissolution profile prediction. To address this challenge, a multi-stage ensemble learning framework is proposed to estimate dissolution behavior at multiple time intervals without replacing regulatory dissolution testing. The proposed framework combines physicochemical descriptors, manufacturing parameters, and dissolution conditions into a unified predictive model. The study emphasizes explainable machine learning to identify formulation variables that most strongly influence dissolution kinetics. Such an approach can support formulation scientists during early-stage product development by reducing experimental iterations. The proposed framework demonstrates the potential of artificial intelligence as a decision-support tool for pharmaceutical quality-by-design (QbD) applications.