<p>Air pollution is recognized as a serious threat to the environment, society, and human health. In Vietnam, air quality has declined in recent years along with urban development, raising concerns about its negative impacts. Addressing this issue requires a comprehensive air quality monitoring and forecasting system to provide timely information, enabling individuals to plan daily activities and take necessary precautions during periods of poor air quality. This paper presents a deep learning-based framework for air pollutant concentrations prediction. The proposed approach utilizes data preprocessing techniques to effectively handle outliers and missing values. Several deep learning models have been developed and evaluated, focusing on data pattern analysis, correlation assessment among air quality indices, and customized input feature selection for each model. The results indicate that two fine-tuned models, LSTM–LSTM and ESN, consistently achieved the lowest prediction errors in most experiments. Additionally, a hybrid model integrating outputs from multiple deep neural networks demonstrated promising accuracy compared to individual models, highlighting its potential for improving air quality forecasting.</p>