Advanced ambient air quality prediction through weighted feature selection and improved reptile search ensemble learning

Air pollution causes a pivotal impact throughout the world that affects natural resources. It also makes hazardous damage to the environment and defects in human health. The World Health Organization states that the report of air pollution is the major reason of human ailments such as lung cancer, early death, asthma, premature birth, and stroke. Due to the influence of changes in weather and climate caused by air pollution, global warming, acid rain, rainfall declines and depletion of the ozone layer occur. To mitigate these issues, preventive measures for air quality are prerequisites. Therefore, air quality monitoring is considered the main aspect of acquiring decision-making support that yields accurate predictions. In addition, there is a need of evaluating the quality of ambient (outdoor) air depending on the observations of pollutants. To achieve this, an automated air quality prediction model is proposed by using modified probability ratio-based RSA (MPR-RSA) and ensemble-based air quality prediction (EAQP). In the first step, the input data are undergone the preprocessing step. The preprocessing is done through various methods such as data imputation, data cleansing, and data transformation. Then, the preprocessed data are given to extract the significant features. The extracted features are obtained by statistical features, spatial features, and temporal features. To enhance the predictive accuracy, the weighted feature selection is employed, where the weight parameter is optimized by the proposed MPR-RSA algorithm. Then, the classification process is accomplished by EAQP, where the hyper-parameters are optimized by the same MPR-RSA algorithm. Here, the ensemble model is constructed by a single Prediction approach as support vector regression, recurrent neural network, extreme learning, bi-directional long short-term memory, and multi-layer perceptron neural network. Finally, the performance is analyzed with various parameters to prove that the proposed model becomes an advanced air quality prediction. Throughout the analysis, the RMSE of the proposed model achieves 9.96%, which can be a lesser value than the other existing heuristic algorithms. Hence, the proposed prediction model attains the low value of RMSE and MAE, which offers early forecasts of ambient air pollution to evade the damage and impacts to the environment. © 2023, The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature.