Low-Cost AirQuality Device
A low-cost outdoor ozone monitoring system built as an accessible alternative to commercial air-quality devices — pairing a MiCS ozone sensor with a novel Adaptive Kalman–LightGBM prediction framework.
What was built
- 01
Designed the AKL/GBM estimator (Adaptive Kalman–LightGBM), a novel ML-based learning and sensor-fusion approach processing real-time data from the Arduino IoT Cloud.
- 02
Implemented the complete software pipeline: data ingestion, preprocessing, adaptive Kalman feedback loops, and LightGBM model training and inference.
- 03
Built evaluation and visualisation frameworks to benchmark performance, generating real-time dashboards and multi-step-ahead prediction metrics (RMSE, MAD).
- 04
Validated performance through large-scale data collection, showing significant improvements over traditional LightGBM approaches in the literature.
- 05
Automated retraining and feedback adaptation, enabling on-premise learning over time and maintaining long-term accuracy across 500K+ samples.