Development of Diagnostic Algorithm Based on Individual Skin Properties, Life Style and Genetic Data for Personalized Solution
Podium 71
Presented by: Eun Bi KO
Customers have used various skin care and functional cosmetics to prevent their skin aging as well as improve current condition of the skin. It is important to analyzing and understanding their skin to select optimized skin care solutions. The aim of this study is to quantitatively analyze the effect of environment, lifestyle, and innate genes on the current skin condition. For this purpose, degree of hydration, sebum, wrinkles, melanin, dullness and redness in a highly controlled condition were collected in conjunction with a questionnaire survey analyzing their lifestyles and genetic data from about 3000 women.
We classified participants based on the types of skin, using 6 kinds of index representing skin properties, questionnaire survey on lifestyle and genetic data with Gaussian Mixture Model and Decision Tree for classification model. Through this study, it was possible to divide the skin of Korean women into 12 clusters according to wrinkles, melanin, redness, dullness and oil/moisture balance with Gaussian Mixture Model. Also, we were able to identify a pattern in which each factor was correlated. For instance, the effect of wearing a mask and fine dust on skin sensitivity was quantitatively analyzed. According to the analysis by decision tree, the biggest influence on aging was the use of sunscreen, and the second was the living habits such as pregnancy or childbirth.
Next, we tried to discover factors for predicting skin condition changes through correlation analysis with lifestyle, climate/environment, and innate genes that affect the current skin condition. We used regression model in order to predict changes in participants’ skin with variables generated during feature engineering. After training the various kinds of regression model, we tried to create a model with good predictive power by applying Ensemble techniques with models which have small RMSE. One of the most interesting parts is the result of analyzing the correlation between genotype and phenotype. Through this, it was possible to understand the effect of acquired lifestyle, environment, and cosmetics in particular on current skin condition.
Ultimate purpose of this study is to predict future skin condition using big data and AI technology. This knowledge will enable us to provide more proactive and personalized solutions not only for cosmetics but also for life care such as lifestyle, eating habits, and environmental response. We will present our initial model to correlate the effect of each factor, including genes, on future skin aging with good predictive power by applying Ensemble techniques with models which have small RMSE.
We classified participants based on the types of skin, using 6 kinds of index representing skin properties, questionnaire survey on lifestyle and genetic data with Gaussian Mixture Model and Decision Tree for classification model. Through this study, it was possible to divide the skin of Korean women into 12 clusters according to wrinkles, melanin, redness, dullness and oil/moisture balance with Gaussian Mixture Model. Also, we were able to identify a pattern in which each factor was correlated. For instance, the effect of wearing a mask and fine dust on skin sensitivity was quantitatively analyzed. According to the analysis by decision tree, the biggest influence on aging was the use of sunscreen, and the second was the living habits such as pregnancy or childbirth.
Next, we tried to discover factors for predicting skin condition changes through correlation analysis with lifestyle, climate/environment, and innate genes that affect the current skin condition. We used regression model in order to predict changes in participants’ skin with variables generated during feature engineering. After training the various kinds of regression model, we tried to create a model with good predictive power by applying Ensemble techniques with models which have small RMSE. One of the most interesting parts is the result of analyzing the correlation between genotype and phenotype. Through this, it was possible to understand the effect of acquired lifestyle, environment, and cosmetics in particular on current skin condition.
Ultimate purpose of this study is to predict future skin condition using big data and AI technology. This knowledge will enable us to provide more proactive and personalized solutions not only for cosmetics but also for life care such as lifestyle, eating habits, and environmental response. We will present our initial model to correlate the effect of each factor, including genes, on future skin aging with good predictive power by applying Ensemble techniques with models which have small RMSE.