Interpretation of gender divergence in consumption places based on machine learning and equilibrium index – A case study of the main urban area of Beijing, China

The influence of gender is often overlooked in spatial studies of consumption places, and it is challenging to track consumers' perceived preferences due to measurement efficiency and scope issues. Based on machine learning techniques, this paper proposes an approach to map perceived preferences from the local to the city level, and evaluate the physical forms and distribution characteristics of gender-specific consumption places. The application steps are: First, acquire POI (point of interest) coordinates of consumption places to obtain street-view images; then, 30 respondents of each gender are invited to score the sampled images. The area proportions of four prominent visual elements in the building façade are identified by Convolutional Neural Network (CNN) and Fully Convolutional Network (FCN), after which Random Forest Model is applied to predict the preference scores of all POIs. Finally, the distribution equity of the consumption places of each gender is evaluated by Equilibrium Index. Applying this method in Beijing, we can identify the gender divergence in four aspects of form, function, scale, and administrative district, locate the area with a poor supply of consumption places, and finally provide optimisation suggestions. The method improves the efficiency and scope of measuring consumers' perceived preferences, while identifying and optimising the quality of consumption places.