Curbside Intensification

Problem

Cities lack data-driven methods to understand how curbside space is actually used throughout the day. Static zoning can't adapt to the dynamic reality of parking, deliveries, pedestrians, and transit competing for the same street space.

Solution

An automated ML system that ingests parking occupancy, pedestrian counts, traffic volume, POI data, and census demographics to classify streets by their daily activity patterns and predict intensification potential.

Pipeline

• Data Ingestion — Automated pipelines pull real-world sensor and administrative data into a unified framework
• Feature Engineering — Street activity aggregated into normalized 24-hour usage profiles with spatial and temporal features
• Unsupervised Learning — K-Means clustering identifies distinct daily activity patterns across urban zones
• Supervised Models — Multinomial logistic regression and random forest predict curbside intensification potential
• Automation — Full CI/CD via GitHub Actions for data fetching, preprocessing, and model execution

Results

K-Means clustering identified distinct daily activity archetypes across urban street segments. Supervised classifiers (logistic regression and random forest) predicted intensification potential with consistent performance across held-out validation sets. The pipeline handles missing sensor data and real-world noise automatically via the CI/CD automation layer.

Stack

Python · Scikit-learn · Pandas · GitHub Actions · GIS Libraries

AIT Austrian Institute of Technology, 2025