Classification of microorganism species using Discriminant Analysis

Abstract

Identification of microorganisms causing root canal infections is an important step in the treatment of these infections. Cultivating the microorganism involved is a relatively difficult and time consuming process. Therefore, clinicians prefer to follow a treatment method based on their prior experience, rather than identifying the related pathogen microorganism and choosing a treatment strategy accordingly. In this study, we have acquired odor data using an electronic-nose equipment with 32 carbon polymer sensors, from pure cultures of 7 microorganisms which are typical causes of root canals infections. We have worked on 28 specimens that are prepared at the Microbiology Laboratory of Pharmacy Faculty. Therefore, there were 4 odor data samples for each of the 7 microorganism types. We have then processed odor data using different pre-processing and dimensions reduction methods and obtained 18 different datasets. We have finally classified these datasets into 7 groups using Discriminant Analysis (DA) and investigated performance of several subtypes of DA algorithm, namely linear, Mahalanobis and quadratic. We have observed that the quadratic approach produces relatively better classification performance. Besides, we have figured out the impact of different pre-processing methods on the classification accuracy.