Using Random Forests on Real-World City Data for Urban Planning in a Visual Semantic Decision Support System

Area of Science:

• Urban Planning and Data Science
• Geographic Information Systems (GIS)
• Machine Learning Applications

Background:

• Increasing urban data volume presents challenges in consolidation, visualization, and exploitation.
• Optimal site selection for commercial or welfare services is a critical urban planning problem.
• Existing buildings and empty spaces require effective utilization strategies.

Purpose of the Study:

• To propose a machine learning approach for addressing urban data challenges.
• To develop a system for combining, fusing, and merging diverse urban data sources.
• To evaluate the effectiveness of machine learning models in urban planning applications.

Main Methods:

• A novel semantic model was developed to encode geometric and semantic urban data.
• Data was fed into a Random Forests classifier and compared with other supervised models.
• Experimental evaluation involved multiple real-world datasets and various performance metrics.

Main Results:

• The proposed system successfully combined and encoded heterogeneous urban data.
• Random Forests demonstrated superior performance across key metrics: Accuracy, Specificity, Precision, Recall, F-measure, and G-mean.
• The approach proved effective in addressing urban planning challenges related to location and space utilization.

Conclusions:

• Machine learning, particularly Random Forests, offers a powerful solution for urban data integration and analysis.
• The developed semantic model enhances the utilization of both low-level and high-level urban data.
• This approach provides a robust framework for data-driven urban planning and decision-making.