Objective Amid China’s strategic push for child-friendly urbanization and its evolving demographic policies, his research explores how urban street environments affect children’s emotional well-being. Focusing on Fuzhou, a national pilot city for child-friendly initiatives, the research addresses a critical gap in urban planning literature: The lack of empirical evidence linking micro-scale street design to the emotional dynamics of children and their caregivers. Existing research primarily prioritizes physical safety and functional infrastructure, while often neglecting the psychosocial dimensions of urban spaces, such as how sensory stimuli, spatial aesthetics, and perceived safety collectively influence residents’ daily emotional states. By examining interactions between street environment elements and residents’ emotional responses, this research aims to generate actionable insights for creating emotionally supportive urban environments that align with China’s child-friendly urbanization goals.

Methods The research employs a multi-modal analytical framework integrating geospatial data, machine learning, and participatory scoring. Data sources include 53,771 Baidu Street View images, 1,474 social media texts (from platforms like Weibo and government portals), and human − machine adversarial scores derived from 40 children − caregiver dyads evaluating street safety perceptions. Three machine learning architectures are deployed: CNN-BiLSTM Hybrid Model, FCN-RF Semantic Segmentation, and XGBoost-SHAP Interpretability Framework. For FCN-RF Semantic Segmentation, street view images are processed by fully convolutional networks to quantify 10 spatial metrics, validated against human-scored safety perceptions via random forest-based adversarial training; for XGBoost-SHAP Interpretability Framework, the nonlinear relationships between 12 street environment indicators and emotional indices are modeled through extreme gradient boosting, with Shapley additive explanations (SHAP) decoding feature contributions and interaction effects. This combination of methods enables detailed analysis of how spatial metrics and perceptions shape emotions.

Results Key findings highlight the nonlinear effects of street environment elements on residents’ emotion. Traffic flow: Moderate traffic flow enhances urban vitality, but excessive traffic flow leads to negative emotion due to noise and safety concerns. SHAP analysis reveals a threshold effect, whereby emotion scores peak and then decline at a given traffic flow: Balanced visual stimuli promote positive emotion, while overly cluttered or monotonous streetscapes reduce emotional satisfaction. Areas such as Academy Road in Gulou District are optimized for visual diversity and exhibit higher emotion scores. Higher safety scores enhance positive emotion, especially in areas with adequate lighting, visible safety facilities, and caregiver-friendly infrastructure. However, poorly maintained security facilities reduce emotional benefits, despite high design scores. For example, in terms of guardrail density, guardrails improve emotion in high-traffic areas, but may create unwelcoming environments that are overly safe in recreational areas, suggesting a dependence on environmental influence. Spatial analysis finds that clusters of low-emotion areas are associated with fragmented pedestrian networks, insufficient green space, and mismatched security measures. Notably, child-friendly renovations in Fuzhou perform poor emotionally due to disjointed maintenance and environmental mismatches, emphasizing the need for adaptive design strategies. In view of this, a three-level optimization path of “traffic control (base layer) — safety creation (middle layer) — spatial quality (enhancement layer)” is proposed.

Conclusion This research advances child-friendly urban planning by street spatial perceptions to residents’ emotional outcomes. Methodologically, the research demonstrates the efficacy of combining machine learning (CNN-BiLSTM, XGBoost) with participatory human − machine scoring. Key practical implications include prioritizing traffic calming measures near schools and residential areas, balancing visual complexity through context-sensitive landscaping to avoid sensory overload or monotony, ensuring that safety infrastructure is supplemented by regular maintenance and caregiver-centered amenities, and employing adaptive fencing strategies that are consistent with spatial functions. Although limited by data granularity and area specificity, this research highlights the importance of embedding sentiment analysis into urban governance. Machine learning and SHAP methodology provide nuanced analysis of how urban environments impact residents’ emotions. These methods not only expand the data base for research on the built environment of child-friendly urban streets, but also validate the feasibility of multi-source fusion of subjective perception data and built environment data in emotion perception measurement, providing an effective methodological reference for the field of spatial research on child-friendly city streets. The present research has made important progress, but there are still limitations in data sources and methods of analyzing residents’ emotions. Future research should expand the diversity of data and refine sentiment recognition models to address cultural and environmental variability. By combining spatial indicators with emotional experiences, this research may contribute to the creation of inclusive, resilient and emotionally supportive child-friendly cities that prioritize safety and well-being.