Air qualityPrediction predictionAir quality prediction (AQP) is critical to reducing the harmful effects of air pollution on the environment and public health. This study offers a thorough method for predicting air quality (AQ) using deep learningDeep learning (DL). The Central Pollution Control Board (CPCB), which is part of the Indian government’s Ministry of Environment, Forests, and Climate Change, provided the air pollution databases used in this study. In order to improve the consistency and quality of the data, this paper used Adaptive Savitzky–Golay FiltersAdaptive Savitzky–Golay filters (ASGF) for preprocessingPreprocessing. These filters were essential in reducing noise and smoothing the data, which guaranteed that the DLDeep learning models were fed high-quality input. It made use of MobileNet, a cutting-edge DL architecture well-known for its effectiveness and efficiency in gleaning relevant features from images and multi-modal data, for feature extraction. This made it easier to transform complicated data on AQ into a format that the suggested model could easily understand. Our new U-Shape Attention-Based Transformer Net (UATNet), which was created to handle the complex spatiotemporal dependencies found in AQ data, was used to carry out the classification process. This paper used virus colony search optimisationVirus colony search optimisation (VCSO) to fine-tune the hyper parameters of the UATNet in order to maximise its performance. With the help of this special optimisation technique, we were able to maximise the model’s predictive power. The suggested VCSO-UATNet model outperforms the current model by demonstrating that it can reach an accuracy rate of 99%, which is an impressive improvement.