Decoding Formulation Complexity: Comparative Performance of Statistical, Machine Learning, and Artificial...

Artificial intelligence (AI) and machine learning (ML) are increasingly explored to improve predictive modelling and optimization in pharmaceutical formulation development. This study compared predictive performance of conventional statistical approaches and AI/ML-based models using an experimentally generated dataset from PLGA-mPEG-based Intelligent Nano Depot (iNDEP), smart depot forming long-acting injectable system for sustained co-delivery of Memantine hydrochloride (MH) and Galantamine hydrobromide (GH). Central Composite Design (CCD) dataset of 17 experimental formulations were employed to develop predictive models. The % cumulative drug release (CDR), implant formation time and particle size were measured to determine effects of PLGA-mPEG, Span 80 and Tween 80. Design of Experiments (DoE), SPSS-based Multiple Linear Regression (MLR), Artificial Neural Networks (ANN) and ML workflows in Orange Data Mining and KNIME Analytics Platform were used to produce predictive models. Root mean squared error (RMSE), mean squared error (MSE), and coefficient of determination (R²) were used to evaluate model performance. Models developed using DoE were good fit with R² values ranging from 0.9882 to 0.9916. The MLR models predicted perfectly (R² = 0.866-0.951) for the %CDR responses using SPSS. ANN models successfully captured nonlinear relationships between formulation and response. Analysis of ML workflows indicated varying performance of algorithms depending on response; there was no one model that performed good across all responses. Traditional statistical and AI-based predictive methods complement each other. The integration of DoE with AI- and ML-based modelling offers a comprehensive tool for enhancing prediction accuracy and guiding formulation and optimization of pharmaceutical products.