Additive Regression Algorithm Predicts The Atmospheric Pollutant Concentrations With Higher Precision

摘要

Due to increased number of deaths related to air pollution, prediction model development has become a key focus for researchers. No doubt, machine learning algorithms based on artificial neural network and support vector machine (SVM), such as multilayer perceptron (MLP), radial base and linear base are the most reliable and widely employed data mining tools for air pollution modelling. However, ensemble learning techniques, such as random forest (RF), bagging, additive regression (AR) have recently emerged as efficient machine learning tools. In view of prediction modelling, the study employs independent classifiers, such as RF, SVM, regression tree using M5 algorithm (M5P) and simple linear regression (SLR) during first phase. Considering the concept of composite modelling, during next phase, all classifiers were combined with AR to assess the prediction performance of each classifier. The study uses emission and meteorological dataset to predict the atmospheric concentration of nitrogen dioxide (NO_2). The prediction performance of each model was evaluated in terms of correlation coefficient (R~2). The model performance was validated by comparing the results of independent classifiers against composite classifiers. The results obtained suggest that adopting independent classifiers within additive regression as base classifiers improve their prediction accuracy and reduce error values.