Article content
Abstract
In response to the problems of subjectivity, evident scale restriction, and lack of dynamism in conventional child-friendly urban environmental evaluations, this paper proposes a machine learning-based multi-dimensional evaluation framework. Based on the central urban regions of the three major Chinese metropolises (Shanghai, Shenzhen, and Guangzhou), it utilizes street view images, remote sensing imagery, POI data, and child behavior trajectory data. Through machine learning algorithms including deep learning and random forest, it performs quantitative evaluations in four aspects: ecological safety, spatial accessibility, facility suitability, and behavioral safety. It is found that machine learning can realize fine-grained extraction of child-friendly environmental indicators with an accuracy rate of 89.7%. Spatially, the quality of child-friendly environments in the three metropolises shows a differentiation pattern that "the core areas are better than the peripheral ones, and the waterfront areas are better than the densely developed ones." The density of recreational facilities and the safety of pedestrians in streets are key elements affecting the evaluation outcomes. This framework offers technical support and experience references for the precise implementation of policies in child-friendly city planning.