Air quality prediction is of critical importance for public health, environmental protection, and urban planning. Traditional methods for air quality prediction often rely on statistical models that may struggle to capture the complex nonlinear relationships inherent in atmospheric dynamics. In recent years, deep learning methods have become effective tools for tackling these challenges. This paper presents an approach of deep learning for air quality prediction, covering three architectures, including LSTM (Long Short-Term Memory), 1DCNN-LSTM (1 Dimensional Convolutional Neural Network-LSTM), and TFT (Temporal Fusion Transformers). The advantages and limitations of these deep learning models as well as the comparison performance among them, are discussed. The experiment result of three deep learning models implemented on the data measurement from Hanoi shows the promising application of deep learning approaches for air quality prediction.