A Data Mining Tool for the Analysis of Epidemiological Data

Abstract

The present paper introduces an integrated approach based on statistical analysis coupled with data mining to analyse epidemiological data. Initially, the statistical properties of the data are analysed. The causality of the exogenous variables (e.g. meteorological and air quality) on the epidemiological data through the Granger causality test is estimated in an attempt to identify those variables that explain major variations. Those variables that are estimated as important are subsequently binned into a finite number of categories as a pre-processing step for the data mining algorithm. The epidemiological and meteorological data are grouped into 5 categories, were as for the air quality parameters the Air Quality Index introduced by U.S. EPA is utilised. Then an algorithm to estimate association rules from the categorised data is developed and applied. The outcomes of the analysis are patterns that relate meteorological and air quality characteristics to specific epidemiological conditions and appear systematically on the examined data set. The application of the developed methodology is performed using data from two major U.S. cities, namely Los Angeles and Pittsburgh.