Send message to Author

Abstract

A growing body of research suggests that students’ academic outcomes are shaped not only by individual and family characteristics but also by the environments in which they learn. However, the role of everyday physical surroundings remains insufficiently measured and understood. This study aims to identify the relationship between the physical characteristics of school environments and student academic performance, using general education schools in Astana as a case study. We propose a methodology that integrates the analysis of Google Street View (GSV) panoramic images with modern vision-language models (VLM). We extract a comprehensive set of interpretable urban environment features—including perceived safety, landscaping, and socio-economic markers—to integrate them with the results of the Unified National Testing (UNT). The methodological novelty of this research lies in the creation of a unified analytical pipeline that integrates computer vision techniques (Qwen3-VL-8B, StreetLens), the extraction of self-supervised embeddings (DINOv2 ViT-L), and quantitative sociological analysis. Unlike traditional studies that rely on aggregated statistics, our approach utilizes high-resolution spatial data to capture the micro-context of the school environment.