Air pollution is becoming a worldwide concern, impacting not only human health but also the environment. Conventionally, monitoring air quality with satellite observations and ground stations proves to be inaccurate for real-time forecasting due to the spatial and temporal resolution shortcomings. With technological advancements in Internet of Things (IoT) and machine learning, sensor-based air pollution forecast models have gained significant attention. Advances in sensor technology and machine learning in recent years have opened the door to improved air pollution prediction models based on real time observations from low-cost sensors. It discusses various predictive models, such as regression-based models, tree-based models, and deep learning models like artificial neural networks (ANNs) and long short term memory (LSTM) networks. The work outlines the merits and demerits of various models and elaborates on issues such as sensor calibration, data pre-processing, and real-time realization

Satellite observations, Ground stations, Real time forecasting, Internet of Things(IoT), ANNs, LSTM

Finally, air pollution forecasting based on sensor data holds great promise for enhancing environmental monitoring and public health interventions. Nevertheless, current challenges like insufficient data merging, missing or incomplete sensor readings, sparse sensor deployment, and the unavailability of long-term models limit the precision and reliability of forecasts. Filling these gaps with state-of-the-art data fusion algorithms, better imputation strategies, better sensor placement, and application of deep learning models will result in more accurate and actionable information. Further, making the models more interpretable with explainable AI will enable better policy adoption by policymakers and environmental organizations

-