A Review On Air Quality Prediction Using Embedded Machine Learning And Deep Learning Models With Quantization Techniques

Abstract: Air pollution is a major global environmental and health concern, necessitating accurate and timely forecasting of pollutant levels to mitigate adverse effects. Recent advances in machine learning and deep learning enable precise air quality prediction, yet deploying these models in real world resource constrained settings remains challenging. Embedded models augmented with quantization techniques offer an efficient solution by reducing computational costs without significant loss of accuracy. This review synthesizes recent developments in air quality prediction using embedded Machine Learning (ML) and Deep Learning (DL) models with emphasis on 8 bit quantization, hybrid architectures, attention mechanisms, and knowledge distillation. The focal study demonstrates a CNN BiGRU model achieving an R² of 0.99 for PM2.5 on IoT devices, balancing high accuracy with model compression. Through analysis of fourteen contemporary works, this paper highlights multi task learning frameworks, spatial temporal modeling over multiple pollutants, and transformer based approaches as emerging trends. Persistent challenges include extending forecast horizons, improving generalizability, and enhancing real time deployment viability. The review concludes by discussing future directions integrating ensemble methods, Bayesian hyperparameter optimization, and quantum learning potentials toward robust, scalable air quality prediction systems.

Keywords: Air Quality Prediction, Deep Learning, Long Short Term Memory (LSTM), Convolutional Neural Network (CNN), Quantization, Edge Computing, Internet Of Things (IoT), Transformer, Knowledge Distillation, Multi Pollutant Forecasting