Machine Learning + Sustainability

Open Networks and Big Data Lab

Problem Motivation

Cities—especially across the Global South—face two intertwined challenges with outsized impact on public health and productivity: degraded air quality and chronic road congestion. High-fidelity sensors and infrastructure-heavy control systems are costly and sparse, leaving large blind spots. Our work closes these gaps using infrastructure-light methods that turn commodity data (phone/CCTV images, sparse air sensors, and coarse mobility datasets) into fine-grained maps and proactive controls that cities can deploy today.

Project Image 1
Project Image 2

Results and Contributions

A. Scalable Urban Air Pollution Monitoring

  1. Vision-based sensing from mobile & CCTV images

    • Trained interpretable ML models that estimate PM2.5 from everyday images (haze as a physically meaningful proxy), enabling real-time, location-aware phone/CCTV sensing.
    • Accuracy: image model MAE 44 µg/m³, improving to 35 µg/m³ via distributed averaging; in-interval accuracy reaches 74% despite noisy labels and limited data.
    • Coverage impact: combined with sensor data, the augmented network revealed 189 previously undetected hotspots, affecting >150,000 residents beyond public monitor coverage.
  2. Interpolation over sparse sensor networks

    • Built space-time interpolation (e.g., Space-Time Kriging) to infer pollution at unmonitored locations/times.
    • Robustness: monthly hotspot prediction achieves 95% precision and 88% recall with 50% simulated sensor failure; tolerant to label noise and device drift.
    • Mechanistic insight: a companion model explains 65% of transient hotspots, highlighting policy-relevant drivers.
  3. Unified city-scale pipeline

    Vision estimates densify spatial coverage; interpolation refines them into calibrated city-wide maps—yielding decision-grade hotspot surfaces from commodity imagery plus a sparse backbone of monitors.

Metric Value
Image MAE → with distributed averaging 44 µg/m³ → 35 µg/m³
In-interval accuracy 74%
Hotspot detection (50% sensor failure) 95% precision / 88% recall
Newly revealed hotspots (residents impacted) 189 (>150k)

B. Understanding & Mitigating Traffic Jams

  1. Characterizing sudden traffic jams

    • Introduced and formally defined “sudden jams” arising from short, intense bursts; constructed “traffic curves” from coarse sources (loop detectors, Uber Movement) across New York City, Nairobi, and São Paulo.
    • Cross-city finding: Nairobi exhibits significantly longer jam durations per road segment than NYC and São Paulo, indicating management gaps and under-utilized capacity.
  2. Reinforcement learning for congestion control (infrastructure-free)

    • Framed freeway management as a centralized, physics-informed RL problem that modulates target speeds to smooth flows without adding infrastructure.
    • Simulation on real road networks shows: +5% throughput, −13% average delay, and up to −5% total stops versus baselines.
    • Integration: jam characterization informs RL policies about where, when, and how aggressively to intervene.
Throughput Average Delay Total stops
+5% -13%% -5%

How do they connect?

Both projects convert low-cost signals into high-value city intelligence. Pollution mapping fuses commodity imagery and sparse sensors into dense, actionable risk surfaces; traffic modeling turns coarse mobility data into proactive control. The common thread is scalable sensing + robust modeling that cities can deploy now.

Members

  1. Ankit Bhardwaj , NYU.
  2. Shiva Iyer, NYU.
  3. Anant Balashankar, NYU.

Publications

Air Pollution

  1. A. Bhardwaj, S. Iyer, Y. Jalan, L. Subramanian. "Learning Pollution Maps from Mobile Phone Images." IJCAI 2022, AI for Good Track. (PDF) (Code)
  2. A. Bhardwaj, A. Balashankar, S. Iyer, N. Soans, A. Sudarshan, R. Pande, L. Subramanian. "Comprehensive Monitoring of Air Pollution Hotspots Using Sparse Sensor Networks." ACM Journal on Computing and Sustainable Societies, 2025. (PDF) (Code)

Traffic

  1. A. Bhardwaj*, S. Iyer*, S. Ramesh, J. White, L. Subramanian. "Understanding Sudden Traffic Jams: From Emergence to Impact. " Development Engineering: Journal of Engineering in Economic Development, 2023. (PDF)
  2. A. Bhardwaj, R. Asim, S. Chauhan, Y. Zaki, L. Subramanian. "Self-Regulating Cars: Automating Traffic Control in Free Flow Road Networks." ACM Journal on Computing and Sustainable Societies, 2025. (PDF) (Code)