Sequential discriminant classification of environments with different levels of exposure to tobacco smoke

Abstract

The use of biomarkers permits the detection of smoking having taken place in an environment. However, no single biomarker is able to differentiate clearly between different types of environments. Multivariate classification models have helped us to differentiate between outdoors, non-smoking indoors, well ventilated smoking indoors, and smoking environments without good air exchange. We found that the variables that enabled us to classify environments most accurately were indoor temperature, 2,5-dimethylfuran and ethyltoluene. A successful prediction rate of 86.5% was obtained by applying both direct fitting and cross validation discriminant (leave-one-out) analyses. Our results show that although a good air exchange ratio decreases the levels of volatile organic compounds in indoor air due to tobacco smoke, significant contamination still remains.