Utilizing artificial intelligence and geospatial analysis to examine the urban built environment, social vulnerability, and child neglect in Los Angeles

BACKGROUND: Child neglect, defined as a parent or guardian's failure to provide basic needs such as food, clothing, shelter, or medical care, is a widespread global public health issue with long-term consequences for child development. OBJECTIVE: To implement an artificial intelligence-based analysis of street-level imagery to detect built and natural environmental characteristics and to associate these indices with neighborhood-level child neglect risk, while controlling for socioeconomic factors. SETTING: Street-level imagery, the Area Deprivation Index (ADI), and child neglect incidents were aggregated to census block groups (CBGs; n = 141) within neighborhoods in Los Angeles, California, USA. METHODS: We analyzed Google Street View images randomly sampled within neighborhoods using semantic segmentation, a computer vision (CV) technique, to quantify environmental features. These measures were aggregated into CBGs and combined into indices representing natural surveillance, natural environment, lighting, and land-use mix. Zero-inflated negative binomial models, incorporating the child population as an offset, were estimated to evaluate associations with neighborhood neglect rates while controlling for the ADI. RESULTS: Higher levels of natural surveillance, natural environment, lighting, and land use mix were each associated with reduced neighborhood neglect risk, controlling for area-level deprivation. CONCLUSIONS: AI analysis of street imagery benefits child welfare research by translating environmental features into measurable indices that reflect residents' subjective perceptions and objective realities. Research can use these tools to improve evidence-based policy, aiding neighborhood strategies and prevention efforts to reduce child neglect.