In vitro dissolution testing plays a key role in controlling the quality and optimizing the formulation of solid dosage pharmaceutical products. Data-driven dissolution models can improve the efficiency of testing: their predictions can act as surrogates to physical experiments and help identify key material attributes / processing parameters that impact product dissolution. Reducing the data (size) requirements of developing such models would significantly improve the utility of dissolution models. In this study, we investigate how Gaussian process regression (GPR) and active learning can reduce the dataset-size requirements for developing predictive models and identifying important processing parameters compared to common practice methods. Initially, we perform a DoE study over five processing parameters and measure the dissolution of compound B to generate a dataset. Using this dataset, we find that GPR provides higher fidelity predictions of dissolution than polynomial models when trained on the same data. In addition, we use Shapley additive explanations to interpret our GPR model and assess processing parameter importance. Through a retrospective analysis, we find that active learning can target a reduced set of experiments (compared to a full DoE) which are particularly conducive for model development and identification of important processing parameters.