Personal Colour between Perceptual Space and Social Practice

Abstract

Personal colour analysis has become a pervasive guide to self-presentation in East Asian beauty cultures, yet its authority rests on a largely qualitative diagnostic practice whose relation to colourimetric structure remains unclear. This study investigates how far contemporary personal colour, as practised by social media influencers, can be modelled within a perceptually uniform colour space. We compiled a dataset of lipstick products from five Japanese Instagram influencers who routinely classify cosmetics into personal colour categories. For each product, the representative colour sample from the manufacturer’s website was converted from sRGB to CIE L*a*b* coordinates and linked to two kinds of label: yellow-base versus blue-base, and the four seasonal types (spring, summer, autumn, winter). Gradient-boosted decision tree classifiers (XGBoost) were trained to predict these labels from L*, a* and b*. The yellow/blue task achieved an accuracy of .82 and macro F1-score of .81, with feature importance dominated by the b* (yellow-blue) component. By contrast, the seasonal task reached only .65 accuracy and .64 macro F1, with lightness L* emerging as the most informative feature and extensive overlap between all four seasons in the a*-b* plane. These findings suggest that influencer practice tracks a perceptually meaningful yellow-blue dimension, while seasonal categories operate as looser narrative constructs that combine lightness and hue in culturally elaborated ways. The study thus positions personal colour as a hybrid formation in which colour-scientific regularities underpin, but do not fully determine, popular regimes of aesthetic classification.

Keywords: Personal Colour, CIE L*a*b*, Statistical Analysis, Machine Learning