An embedded machine learning system method for air pollution monitoring and control

Air pollution has very serious health impacts as well as detrimental effects on ecosystems. Addressing this issue will require operational real-time monitoring and management, specifically in urban environments. Potential sensors that could provide the necessary high accuracy are too costly for mass deployment in the Internet of Things (IoT) environment. On the other hand, low-cost sensors usually have noisy, unreliable, and less accurate outputs, preventing them from being effective tools in real-time applications. Therefore, there is a lack of cost-effective, accurate, and scalable solutions to enhance the reliability of low-cost sensors such that they can improve air quality monitoring and control in real-time. Consequently, this research proposes a novel approach that advances the use of embedded machine learning with edge computing to enhance the accuracy and real-time actions at the local level. In the system, the Raspberry Pi was utilized for processing sensor information as the embedded edge device to monitor multiple real-time air pollutants, including SO2, NO2, and PM2.5. In this research, the one-rank cuckoo search-driven adaptive support vector machine (ORCS-ASVM) was proposed to improve the prediction accuracy of low-cost sensors with noise. The ORCS-ASVM machine learning model predicts air quality and reduces the amount of computing power required while providing immediate knowledge about the amount of air pollution present. This model strives to enhance the accuracy of low-cost sensor readings, thus accommodating erroneous sensor readings in pollution prediction. Data on air pollution that is continuously collected from the sensors will be monitored at regular intervals to provide real-time monitoring. Sensor data often contains noise, requiring filtering and preprocessing techniques to correct sensor errors. Smoothing and normalization are the most common preprocessing methods used to correct sensor errors. Compared with traditional prediction algorithms, the proposed model outperformed traditional methods in terms of error reduction. Air pollutants such as PM2.5, SO2, and NO2 achieved detection accuracies of 94.2%, 95.2%, and 94.8%, respectively. An embedded machine learning system approach for air quality monitoring and in situ pollution mitigation is a low-cost method that can be extended to other industries, making it suitable for smart city infrastructure, smart agriculture, and other important areas. An embedded machine learning approach is applicable to many areas as it has great scalability and flexibility and is therefore very suitable for widespread adoption by IoT-based pollution monitoring networks.