Lightweight Edge AI framework for Real Time Air Quality Analysis

Recent developments in the field of machine learning and deep learning have led to the improvement of air quality monitoring and forecasting, which allows making more accurate, efficient, and scalable predictions.Most of the current solutions are based on large models that require significant computing capabilities or continuous connectivity with a cloud, which limits their use in resource-constrained environments. The model presented in this paper is a lightweight Gated Recurrent Unit (GRU) that was trained on one year of historical data (365 days) of the Open-Meteo API and then connected to low-cost IoT sensors (PMS7003 to measure PM2.5/PM10 and DHT22 to measure temperature/humidity) on a microcontroller (ESP32). With the help of TensorFlow Lite (TFLite) to infer the model on-the-fly, the model demonstrates high performance with an R2 of 0.9402, Mean Absolute Error (MAE) of 0.0367 and Root Mean Square Error (RMSE) of 0.0424 and has minimal resource needs in relation to the more traditional methods. The proposed system provides a high level of efficiency and edge deployment practicable than traditional benchmarks and is most appropriate to mobile platforms and resource constrained platforms. The study offers a convenient, practical system of real time air quality intelligence in emerging economies.