Skin product formulation development remains resource-intensive because formulation performance depends on interacting ingredients, release and permeation behaviour, and slow experimental evaluation. In vitro permeation testing (IVPT), the standard method for assessing topical and transdermal performance, can require many hours per run and is limited by diffusion-cell capacity. This thesis by publication develops a data-driven optimisation framework to improve decision-making across this workflow.

A central novelty of the thesis is to treat formulation acceleration as a set of linked decisions about which formulations to evaluate, how long to run each experiment, and which test to use, rather than as separate problems. The thesis develops two new methods for expensive experimental campaigns: confidence-bound early stopping with sequential calibration (CBES), which makes experiment duration a controllable decision while limiting the risk of stopping a promising candidate; and soft-rejection-informed Bayesian optimisation (SRBO), which reuses the incomplete data from stopped experiments instead of discarding them.

The thesis starts from a literature review of optimisation methods for skin product formulation that identifies the need for more adaptive, uncertainty-aware workflows, followed by four connected studies supporting the framework. First, Bayesian optimisation is compared with response surface methodology for ibuprofen-loaded poloxamer 407 formulations, showing improved formulation selection in an IVPT-based case study. Second, a confidence-bound early stopping approach is developed to reduce the unnecessary continuation of long experiments while controlling the risk of falsely stopping promising candidates. Third, early stopping is integrated with Bayesian optimisation through a soft-rejection mechanism, so that partial trajectories from stopped experiments can still inform future candidate selection. Finally, in vitro release testing (IVRT) and IVPT are compared to assess whether faster release testing can support upstream formulation screening.

Demonstrated on an ibuprofen-poloxamer formulation system, the framework provides practical tools for more resource-efficient, data-driven formulation development.