Advancing air quality prediction models in urban India: a deep learning approach integrating DCNN and LSTM architectures for AQI time-series classification

Every year, a large number of people in India lose their lives as a result of extreme degradation of urban air quality. This study introduces a deep learning architecture designed to analyze and forecast outdoor air conditions in the Indian capital, Delhi. Meteorological and air quality data are sourced from monitoring stations at fourteen locations across Delhi, focusing on carbon monoxide (CO) and particulate matter (PM_2.5 and PM₁₀) concentrations to derive the Air Quality Index (AQI) time series spanning September 2014–July 2020. The extensive 1765-day AQI dataset is divided into training (1412 days, 80% of the total dataset) and testing (353 days, the remaining 20%) sets. The AQI values, categorized into five classes ranging from satisfactory to severe, serve as the basis for training and testing the proposed models. The development of deep convolutional neural networks (DCNN) and combined DCNN-long short-term memory (DCNN-LSTM) architectures incorporates meteorological parameters such as temperature, humidity, atmospheric pressure, and AQI time-series data. While deep convolutional layers excel in extracting valuable information and discerning the internal structure of input time series, long short-term memory (LSTM) networks are adept at discerning long-term and short-term correlations. The DCNN-LSTM architecture is proposed for its effective integration of the strengths of both DCNN and LSTM architectures. Evaluation metrics including sensitivity, specificity, accuracy, F1 score, and the AUC-ROC curve gauge the predictive performance of the architectures. The DCNN network attains an overall accuracy of 94.48%, an F1 score of 97%, and an AUC of 0.94. Notably, the DCNN-LSTM architecture surpasses other predictive models, achieving the highest classification accuracy of 97.48%, an AUC of 0.97, and an overall F1 score of 97.48%. These results underscore the efficacy of the proposed deep learning approach in advancing air quality prediction models.