Sci Rep. 2025 Aug 6;15(1):28719. doi: 10.1038/s41598-025-09508-8.
ABSTRACT
With the instantaneous economic development, air quality keeps on dwindling. Some key factors for the emergence and evolution of air pollution are high-intensity pollution emissions and adverse weather circumstances. In air pollutants, Particulate Matter (PM) possessing less than 2.5Mu is considered the most severe health issue, resulting in respiratory tract illness and cardiovascular disease. Therefore, it is mandatory to predict PM 2.5 concentrations accurately to ward off the general public from the desperate influence of air pollution in advance owing to its complex nature. Aiming at the complexity of air quality prediction, a new method called Hebbian Concordance and Attention-based Long Short-Term Memory (HC-ALSTM) is proposed. The HC-ALSTM method is split into four sections. They are preprocessing using the Statistical Normalization-based Preprocessing model, feature extraction employing the Generalised Hebbian Spatio Temporal Feature extraction model, feature selection using Concordance Correlation function, and Attention-based Long Short-Term Memory for air quality prediction. First, the Statistical Normalization-based Preprocessing model is applied to the raw dataset to normalize the impact of distinct air pollutants on the bordering factor. Second, with the Generalised Hebbian Spatio Temporal Feature extraction algorithm, processed samples are applied to extract the dimensionality-reduced spatio-temporal feature. Third, with the extracted features, essential or significant features are selected using Concordance Correlation analysis that determines the impact of pollutant concentration of bordering places for predicting air quality index involving both city and state, daily and hourly. Finally, Attention-based Long Short-Term Memory is applied to the extracted and selected features to predict air quality accurately. Through evaluation and analysis using two other evaluation methods, the proposed HC-ALSTM method performs better in error and time. Our method has dramatically improved air quality prediction accuracy and overhead compared with other methods.
PMID:40770205 | DOI:10.1038/s41598-025-09508-8
