Attentive Multi-Task Prediction of Atmospheric Particulate Matter: Effect of the COVID-19 Pandemic

Abstract

Air pollution, especially the continual increase in atmospheric particulate matter (PM), is a global environmental challenge. To reduce the PM concentration, a remarkable amount of machine learningbased research has been proposed. However, increasing the accuracy of the predictions and providing clear interpretations of the predictions are challenging. In particular, no studies have addressed models that predict and interpret PM before and during the COVID-19 pandemic. In this paper, we present a two-step predictive and explainable model to obtain insights into reducing PM. We first use attentive multi-task learning to predict the air quality of cities. To accurately predict the concentration of particles with sizes of ∼10 µm or ≤2.5 µm (PM 10 and PM 2.5 , respectively), we demonstrate a performance difference between single-task and multi-task learning, as well as among the state-of-the art methods. The proposed attentive model with multi-task learning outperformed the others in terms of accuracy performance. We then used Shapley additive explanations, a representative explainable artificial intelligence framework, to interpret and determine the significance of features for predicting PM 10 and PM 2.5 . We demonstrated the superiority of the proposed approach in predicting and explaining both PM 10 and PM 2.5 concentrations, and observed a statistically significant difference in air pollution before and during the COVID-19 pandemic.

Resolving contributions of NO2 and SO2 to PM2.5 and O3 pollutions in the North China Plain via multi-task learning

“…MTL model is trained to perform multiple tasks simultaneously to improve the generalization performance by leveraging the information shared across tasks 32 – 34 MTL trains neural networks to perform multiple tasks by sharing certain layers and parameters of the network between tasks. By sharing some of the network’s parameters, the model can learn a more efficient and compact representation of the data, which can be beneficial when the tasks are related or have some commonalities.…”

“…25,31 Multi-task learning (MTL), a machine learning technique, can be trained to perform multiple tasks simultaneously. [32][33][34] MTL trains neural networks to perform multiple tasks by sharing certain layers and parameters of the network between tasks; the model can learn a more efficient and compact representation of the data, which can be beneficial when the tasks are related or have some commonalities. Considering that PM 2.…”

“…Multi-task learning (MTL), a machine learning technique, can be trained to perform multiple tasks simultaneously 32 – 34 MTL trains neural networks to perform multiple tasks by sharing certain layers and parameters of the network between tasks; the model can learn a more efficient and compact representation of the data, which can be beneficial when the tasks are related or have some commonalities. Considering that PM2.5 and normalO3 pollution have some common influencing factors (including some common precursors), the MTL model is appropriate for simultaneously simulating PM2.5 and normalO3 pollution in the NCP.…”

Interpretable Machine Learning Approaches for Forecasting and Predicting Air Pollution: A Systematic Review

“…There are several methods for visualizing and understanding the contribution of each feature to a prediction. Gu et al (2021), Song et al (2022), andWu et al (2022) have used SHAP explanation force plots to evaluate the significance of features for the prediction of NO2, PM10, and PM2.5 respectively. In these plots, each Shapley value is a force that either increases (positive value) or decreases (negative value) the prediction.…”

“…We can analyze the entire model by studying this matrix through various methods, such as the SHAP feature importance plot. Another method is the SHAP summary plot was used to combine feature importance with feature effects (Alvarez and Smith, 2021;Gu et al, 2022;Han et al, 2022;Kang et al, 2021;Kim et al, 2021;Marvin et al, 2022;Nabavi et al, 2021;Ren et al, 2022;Song et al, 2022;Wei et al, 2022;Wu et al, 2022). This plot can give insights into the relationship between the feature values and their impact on the prediction in each instance.…”