Low-Cost Air Quality Forecasting in Resource-Constrained Environment

Abstract

Rapid urbanization exacerbates air pollution in
developing regions, yet monitoring remains limited by high
costs and infrastructure gaps. This paper presents an IoTedge
framework integrating Raspberry Pi Pico, ESP32-WROOM
microcontrollers, and Plantower PMS7003 sensors for real-time
PM2.5/PM10 monitoring and 7-day AQI forecasting. Leveraging
lightweight SVR regression trained on localized meteorological
data, our system achieves MAE=2.8 for 48-hour predictions while
operating at <10W power. The system enables community-scale
deployment validated in Kathmandu Valley, demonstrating 92%
uptime with component costs under $35/unit. The architecture
addresses key limitations of cloud-dependent systems through
edge processing, making it viable for low-connectivity regions.