Predictive model for the spreadability of cosmetic formulations based on large amplitude oscillatory shear (LAOS) and machine learning

Abstract

Inspired by the analogy between the application process of cosmetics and large amplitude oscillatory shear (LAOS), we suggest a novel predictive model for the spreadability of cosmetic formulations via LAOS analysis and machine learning techniques. Rheological measurements of cosmetic formulations, including the transient elastic and viscous moduli from the sequence of physical process (SPP) analysis, were selected as features for the predictive models, and the spreadability of each formulation that is quantitatively rated by trained panels was set up as the target variable. First, multiple linear regression prediction models are derived, and it was shown that the LAOS-SPP parameters were more effective features than other rheological parameters that were conventionally related to spreadability of cosmetics. Additionally, a non-linear prediction model was developed based on the random forest regressor algorithm, considering the possibility of the nonlinear correlation between rheological measurements and spreadability. The random forest regressor model showed better performance than the linear regression model, and the LAOS-SPP parameters were found to be more effective features for the random forest regressor model as in the multiple linear regression model. The correlation between the LAOS-SPP parameters and the spreadability is interpreted in terms of the rheological transition during rubbing process of cosmetics. Our findings indicate the importance of the nonlinear rheological behavior in the texture perception mechanism of cosmetics, and how rheological measurements can be combined with machine learning techniques to solve complicated problems.