TREND ANALYSIS OF STREAMING AIR QUALITY DATA OF FOUR TIER I CITIES IN INDIA

Abstract

Time series data mining refers to the process of extracting relevant information from time series data. Trend analysis of time series data refers to a set of techniques which help to reveal hidden patterns in time series data and it is important as it helps to make prediction based on past data. Time series mining finds its applications in a number of fields like medical, finance, weather forecasting and many more. In this report trend analysis of air quality data has been discussed. Air pollution has been a major concern of study as it can have adverse impact on human health as well as ecosystem. Not much research has been carried out on air quality data of India. The cities in India have been classified into Tier I, Tier II and Tier III on the basis of their population by Reserve Bank of India. This report aims at doing the trend analysis of air quality of Tier I cities in India. The Tier I cities included are New Delhi, Mumbai, Chennai and Bengaluru. The air quality data included parameters are SO2, NO2, NOx, NO, CO, O3, temperature, relative humidity, PM2.5 and PM10. The air quality data collected has a large number of dimensions and hence to reduce dimensionality, after a number of experiments, principal component analysis has been found to be the best technique for dimensionality reduction to reduce the number of variables under consideration. The results of interpretation of principal component analysis have been used to provide a useful description that can be interpreted in terms of sources of air pollution. After removing seasonality from air quality data ARIMA model has been applied to better understand the data and also predict future values in the time series data. GARCH model has also been applied on the data and the results of ARIMA and GARCH model are quite comparable. In previous studies, ARIMA has been applied only on static data but in this study SDA (Streaming Data ARIMA) has been proposed that applies ARIMA model on streaming data and estimates how various parameters of ARIMA model change with every iteration based on window size of streaming data. The results of SDA and ARIMA on static time sequence are compared and have been found to be very promising.