PM$_{2.5}$ Forecasting
Welcome to the forecasting hub! This section chronicles our complete journey building operational air quality forecasts for Hanoi and Ho Chi Minh City (HCMC).
We set out to see if we could build a highly accurate, 72-hour forecast using purely accessible meteorology—and push the boundaries of what is possible without expensive Chemical Transport Models. Read through our process, our successes, and our failures below!
The Journey
- Literature Review — A practitioner-oriented review of forecast methods, and why the Southeast Asian tropical climate presents a unique, under-studied challenge.
- Statistical Modeling — How we bridged the gap between heavy CTMs and Statistical ML, detailing the “Predictability Ceiling” and our Delta-Skip MLP architecture.
- Hanoi & HCMC Experiments (2026) — The full lab notebook. See how our models performed across different feature sets (V2, V3, V4), why Temporal features failed catastrophically, and the dramatic difference between forecasting Hanoi vs HCMC.
- Results vs Literature — Positioning our final metrics against the global research literature.
- Live Demo — See the final deployed models running live against the latest sensor data for both cities!