A generalizable ensemble meta-learning framework for assessing 15-minute cities.

Area of Science:

• Urban Planning and Sustainability
• Artificial Intelligence in Urban Studies
• Geographic Information Systems

Background:

• The 15-Minute City concept is crucial for urban resilience and sustainability amidst climate change.
• Existing assessment methods for 15-Minute Cities are fragmented and lack a comprehensive, data-driven approach.
• A need exists for a systematic, multi-dimensional framework to evaluate city performance within this concept.

Purpose of the Study:

• To develop a robust, scalable, and comparable assessment model for 15-Minute Cities using ensemble meta-learning.
• To address the limitations of current fragmented assessment approaches.
• To provide a practical tool for evidence-based urban planning and policy development.

Main Methods:

• An ensemble meta-learning model was developed using eight distinct machine learning techniques.
• The model was trained and evaluated using data from 21 cities across five continents.
• The framework was validated and applied to districts in Istanbul, comparing performance with existing literature.

Main Results:

• The ensemble meta-learning framework achieved an overall accuracy of 78.5%.
• F1 scores reached 79.1% for compliant areas and 77.8% for non-compliant areas.
• Performance was robust across diverse contexts, though lower in areas with limited open data.

Conclusions:

• The proposed framework offers a scalable, reproducible, and transparent method for 15-Minute City assessment.
• It combines cross-continental transferability with interpretable spatial outputs.
• This provides a practical tool for informed urban planning and policy decisions.