Smoking status classification by optical spectroscopy and partial least square regression

Abstract

Abstract Smoking status of individuals is often revealed through self-reported data and interviews. The incidence of false reports severely impairs the proper assessment of the individuals’ health conditions and their risk to tobacco associated diseases, delays clinical intervention and treatment services. This paper presents the use of optical technique combined with partial least square (PLS) regression model in the classification of smoking status. The focus of this work is on light absorbance signals (by skin tissue) in the wavelength range of 520 – 600 nm; based on these data dendogram and PLS techniques are used to estimate the number of likely clusters within the considered dataset and to predict smoking status of individuals, respectively. The results from the processing of spectral information of smoking and nonsmoking populations revealed a high misclassification rate of 26.67 % using dendrogram method, but a considerably high accuracy of 90 % evaluated via leave one out cross validation was obtained using PLS component number 4. This study concluded that the spectral oscillation patterns and descending rates corresponded to nonsmoking and smoking individuals could be differentiated and specified using PLS technique in the determination of smoking status.