People stink!: Towards identification of people from breath samples

Abstract

The paper addresses the potential to use breath samples for identifying people. Participants were asked to exhale ten times for a length of five seconds to a tube attached to a commercial ion-mobility spectrometry device on three separate sessions. The data of each participant was divided into training (50% of the samples) and test data sets (50% of the samples) in random order. Classification decision tree (CDT), K nearest neighbor (KNN), naïve Bayes (NB), linear discriminant analysis (LDA), and quadratic discriminant analysis (QDA) were used to analyze if the data could be classified correctly. Within a session, KNN (75.2%), NB (78.3%), and LDA (85.8%) were able to identify participants. Between sessions, the performance decreased.