# An IoT enabled system for enhanced air quality monitoring and prediction on the edge

## TL;DR

This paper introduces an IoT-based system that predicts PM2.5 air pollution using machine learning on edge devices and the cloud, improving air quality monitoring in remote areas.

## Contribution

The novel contribution is an IoT system using hybrid ML algorithms on edge devices and cloud for accurate and fast PM2.5 prediction.

## Key findings

- NARX/LSTM achieved highest accuracy in PM2.5 prediction with lowest error metrics.
- NARX/XGBRF provided best balance of accuracy and speed on Raspberry Pi edge devices.
- The system enables real-time air quality monitoring in low-bandwidth or offline environments.

## Abstract

Air pollution is a major issue resulting from the excessive use of conventional energy sources in developing countries and worldwide. Particulate Matter less than 2.5 µm in diameter (PM2.5) is the most dangerous air pollutant invading the human respiratory system and causing lung and heart diseases. Therefore, innovative air pollution forecasting methods and systems are required to reduce such risk. To that end, this paper proposes an Internet of Things (IoT) enabled system for monitoring and predicting PM2.5 concentration on both edge devices and the cloud. This system employs a hybrid prediction architecture using several Machine Learning (ML) algorithms hosted by Nonlinear AutoRegression with eXogenous input (NARX). It uses the past 24 h of PM2.5, cumulated wind speed and cumulated rain hours to predict the next hour of PM2.5. This system was tested on a PC to evaluate cloud prediction and a Raspberry Pi to evaluate edge devices’ prediction. Such a system is essential, responding quickly to air pollution in remote areas with low bandwidth or no internet connection. The performance of our system was assessed using Root Mean Square Error (RMSE), Normalized Root Mean Square Error (NRMSE), coefficient of determination (R2), Index of Agreement (IA), and duration in seconds. The obtained results highlighted that NARX/LSTM achieved the highest R2 and IA and the least RMSE and NRMSE, outperforming other previously proposed deep learning hybrid algorithms. In contrast, NARX/XGBRF achieved the best balance between accuracy and speed on the Raspberry Pi.