Classifying Severity of Unhealthy Air Pollution Events in Malaysia: A Decision Tree Model

Abstract

The application of data mining technique in dealing with real problems is popular and ubiquitous in various knowledge domains. This study proposes the concept of severity measures correspond to the characteristics of duration and intensity size for evaluating unhealthy air pollution events. In parallel with that, the present study also proposes a decision tree as a predictive model to deal with a binary classification corresponding to extreme and non-extreme unhealthy air pollution events, which is established based on threshold of the power-law behavior. In a similar vein, other characteristics, such as duration and intensity size, were also determined as important related features. A case study was conducted using the air pollution index data of Klang, Malaysia, from January 1st, 1997 to August 31st, 2020. The results found that the decision tree model can provide a high degree of precision and generalization with 100% accuracy in classifying a class for extreme and non-extreme events for the air pollution severity in the Klang area. In addition, a duration size is the most influential feature that leads to the occurrence of an extreme air pollution event. Thus, this study also suggests that authorities should exercise some vigilance precautions with respect to pollution incidents with a consecutive duration exceeding 11 hours.