An IoT–Machine Learning Proof-of-Concept Framework for Real-Time Urban Particulate Matter Prediction Using Low-Cost Sensors

Public health and sustainable urban management continue to face significant difficulties from particulate matter (PM) pollution, particularly PM1.0, PM2.5, and PM10. However, low-cost real-time sensing technologies appropriate for localized deployments have not been thoroughly investigated. A proof-of-concept IoT-machine learning framework that combines regression-based prediction with clustering-based exposure pattern analysis is presented in this paper, which tackles the issue of real-time particle monitoring in a campus setting. A PMS7003 dust sensor interfaced with an Arduino module is used for data collecting. Preprocessing procedures such as outlier treatment, normalization, and temporal indexing are then carried out. To represent nonlinear temporal behavior, random forest and polynomial regression models are used, with predictive performance for PM1.0and PM2.5surpassing 85%. In the monitoring region, K-means clustering is also employed to find high-exposure zones that are both emerging and persistent. Due to discretized and low-variability sensor outputs, performance declines for higher particle sizes (5.0 μm and 10.0 μm), highlighting crucial data quality restrictions for efficient regression modeling. Performance measurements including mean squared error, mean absolute error, and the coefficient of determination (R2) are used to support the analysis. The suggested framework facilitates proactive exposure mitigation techniques that are in line with the Sustainable Development Goals for resilient infrastructure, health, and climate action. It also provides a scalable method for real-time environmental monitoring. The thorough documenting of performance trade-offs across various particle sizes and the combined application of supervised and unsupervised learning approaches for campus-scale air quality assessment constitute the main contributions of this work.