Prediction of sensory textures of cosmetics using large amplitude oscillatory shear and extensional rheology

We propose a predictive model for various sensory textures utilizing machine learning techniques based on the largest rheology and panel-tested sensory texture database ever assembled. In addition to the conventional rheological parameters typically measured in the cosmetics field, rheological parameters obtained from the large amplitude oscillatory shear (LAOS) sequence of physical processes (SPPs) and extensional rheology analyses are employed as feature variables for the predictive model. These feature variables are chosen to mimic real flow conditions during the usage of cosmetics, such as rubbing and tapping, as they are expected to contain more information related to sensory textures. It has been demonstrated that our prediction model, based on the random forest regression algorithm, can effectively predict five sensory textures: spreadability, thickness, softness, adhesiveness, and stickiness. We investigated the rheological characteristics crucial for determining each sensory texture through permutation and feature importance analyses. The important analysis highlighted the close correlation between rheological parameters from LAOS–SPP, extensional analyses, and sensory textures. By using this correlation, we interpret the perception of each sensory texture in the context of rheology.