Implementation of a Low-Cost Air Quality Monitoring System Using Neural Network Forecasting Model

Abstract

This study presents the development of a low-cost air quality monitoring system designed to assess the Air Quality Index (AQI) in Port Harcourt, Nigeria, using a neural network-based prediction model. The system integrates affordable environmental sensors to gather real-time data on pollutants such as PM2.5, PM10, NO2, SO2, and CO2, as well as environmental parameters including temperature and humidity. These sensors interface with Arduino-based microcontrollers, and data is logged using SD card modules for further processing. The predictive aspect of the system is powered by a Long Short-Term Memory (LSTM) neural network model trained on historical air quality and meteorological data to improve forecast accuracy. The model’s performance, evaluated using Mean Squared Error (MSE), achieved a low training loss of 0.0189 and a validation loss of 0.00067394, indicating high precision in AQI predictions. The results show that the LSTM model significantly outperforms traditional prediction models and earlier neural network-based approaches, particularly in the accuracy of PM2.5 and PM10 forecasts. The low-cost sensors used in the system demonstrated strong agreement with reference-grade air monitoring equipment, especially in tracking particulate matter levels. PM2.5 and PM10 predictions closely followed World Health Organisation (WHO) standards, aligning with recommended mean limits for air quality safety. Additionally, the affordability of the system is notable; the prototype costs only 9% of a lower-end commercial device and 0.45% of a higher-end system, enhancing accessibility for broader deployment. This makes the solution highly scalable and practical for both urban and rural environments. Overall, the project contributes a robust, cost-effective, and accurate air quality monitoring solution that leverages AI for real-time prediction, offering significant implications for environmental monitoring, public health protection, and data-driven policy making.