Machine Learning enhanced e-tongue for the characterization of a new paracetamol-based formulation for pediatric delivery

The development of palatable pediatric formulations of paracetamol remains a major pharmaceutical challenge due to its intrinsic bitter taste and the limited availability of well accepted solid dosage forms for children. This work presents a machine learning enhanced electronic tongue platform for the characterization of two newly developed dextrin based granulated formulations designed for pediatric delivery. A portable and wireless dual channel potentiostat was employed to acquire differential pulse voltammetry measurements from gold nanoparticle modified screen printed electrodes. A dataset including pure paracetamol, blank excipient formulations, and drug loaded granules at two concentration levels were collected and processed using four machine learning classifiers, namely Support Vector Machines, Kernel Logistic Regression, k Nearest Neighbors, and an ensemble tree method. Model performance was evaluated in terms of accuracy, F1 score, prediction speed, and implementation footprint. Support Vector Machines and Kernel Logistic Regression achieved the highest validation performance, while ensemble and k Nearest Neighbors models provided reduced computational complexity at the expense of accuracy. Importantly, all trained models consistently classified the new granulated formulations as blank samples rather than pure paracetamol, demonstrating effective taste masking capability. The results confirm the potential of combining portable electrochemical sensing and machine learning to support rapid and objective assessment of organoleptic related properties in pediatric pharmaceutical development.