PharmaQAI: A Machine Learning Framework and Interactive Platform for Process Intelligence and Quality Prediction in Pharmaceutical Tablet Compression

Predicting pharmaceutical product quality from manufacturing process parameters is a central objective of Quality by Design and Process Analytical Technology frameworks. This study presents a systematic machine learning analysis of 1,005 tablet compression batches characterized by 27 process parameters and six critical quality attributes including drug release, residual solvent, and total impurities. Nine regression and seven classification algorithms were evaluated with randomized hyperparameter optimization and five-fold cross-validation. Tree-based ensemble methods, particularly Extra Trees, consistently outperformed linear approaches across all quality targets. Total impurities achieved the highest predictive accuracy with a test R² of 0.8855, driven primarily by formulation-specific categorical identifiers, while drug release targets yielded moderate R² values of 0.40 to 0.47, reflecting the inherently complex process-dissolution relationships. Classification of weekend batch production using Logistic Regression yielded an AUC of 0.9215 and cross-validated accuracy of 0.9493, confirming that production schedule characteristics are reliably encoded in process signatures. Feature importance analysis identified tablet fill weight and compaction force as the dominant drivers of dissolution performance. A dedicated Streamlit web application, PharmaQAI, was developed to integrate data exploration, supervised model training, residual diagnostics, and interactive contour-based response surface visualization into a single accessible platform, supporting proactive data-driven decision making in pharmaceutical manufacturing without requiring programming expertise.