click here to download the base paper
Abstract
Air quality has a significant impact on human health. Degradation in air quality leads to a wide range of health issues, especially in children. The ability to predict air quality enables the government and other concerned organizations to take necessary steps to shield the most vulnerable, from being exposed to the air with hazardous quality. Traditional approaches to this task have very limited success because of a lack of access of such methods to sufficient longitudinal data. In this paper, we use a Support Vector Regression (SVR) model to forecast the levels of various pollutants and the air quality index, using archive pollution data made publicly available by Central Pollution Control Board and the US Embassy in New Delhi. Among the tested methods, a Radial Basis Function (RBF) kernel produced the best results with SVR. According to our experiments, using the whole range of available variables produced better results than using features selected by principal component analysis. The model predicts levels of various pollutants, like, sulfur dioxide, carbon monoxide, nitrogen dioxide, particulate matter 2.5, and ground-level ozone, as well as the Air Quality Index (AQI), at an accuracy of 93.4 percent.
Keywords—air quality index, support vector regression, radial basis function
INTRODUCTION
The sharp rise in air pollution in recent years, due to industrial and agricultural activities, as well as increased number of vehicles using internal combustion engines, has caught the attention of the scientific community . Air pollution has significant impact on human health and may cause long-term health issues in children. The significant rise in air pollution in New Delhi is attributed to increased vehicular emissions, burning of fossil fuels at power plants, and other local industries and burning of fields by farmers in
neighboring states .Air quality is being monitored in New Delhi for about two decades. This has allowed a better understanding of the changes in air pollution in response to particular activities and government regulations, but the air pollution in New Delhi remains a problem.Air pollution is responsible for 30 percent of lowerrespiratory tract infections and is linked with 91percent of premature deaths, from lung cancer, heart disease, acute respiratory infections, stroke and chronic obstructive pulmonary disease. It contributes to 20 percent of infant mortality worldwide and causes numerous short- and longterm illnesses in children. Exposure of the mother to high levels of air pollution can lead to adversely affect immune status, brain development, respiratory systems, and cardiometabolic health of the child. Air pollution has also been linked to low birth weight and stunted growth in children. Air pollution is estimated to be responsible for one in ten deaths of children under five years of age. In elder people,
air pollution causes high rates of asthma, with decreased cognitive performance.Presently, the government implements regulations after the air quality reaches hazardous levels. If there is a way to
foresee the air quality reaching hazardous levels, the government can implement such regulations early,
potentially preventing further degradation of air quality and being able to shield those, most vulnerable, from getting exposed to such air quality. This study aims to build a model that can look at previously recorded air quality data and predicts levels of different pollutants as well as air quality index. For this we use a variation of Support Vector Machines (SVM), called Support Vector Regression (SVR). The paper is organized as follows. We state the motivations of this work and frame our work in section 2, stating the potential impact of being able to successfully predict the air quality. We provide a critical revision of related work, done previously, in Section 3. We explain how Support Vector Machines (SVM), particularly Support Vector Regression works in Section 4. We describe the datasets used in this work and the data preprocessing steps used to produce more efficient input for the SVR, in sections 5 and 6 respectively. In Section 7, we present details of the experiment performed, divided into subsections describing the experimental setup and the results obtained. In Section 8, we conclude the paper and discuss ideas for future work in this area.
