Applying the Em-algorithm to Classiication of Bacteria Mats Gyllenberg Applying the Em-algorithm to Classiication of Bacteria

In present paper we study the use of the expectation maximization (EM) algorithm in classi cation. The EM-algorithm is used to calculate the probability of each vector belonging to each class. If we assign each vector to the class of maximal probability we get a classi cation minimizing a certain log-likelihood function. By analyzing these probabilities we get a clearer picture of how well data ts to the classi cation than by traditional classi cation methods. We de ne a vector to be well classi ed in the classi cation if its probability of belonging to some class is above a prescribed value 1 . Then we set up the experimental procedure to lter out elements that are not well classi ed in a large data set describing strains of bacteria belonging to the family Enterobacteriaceae. We compare classi cations with subset of the data (containing only well classi ed elements) to classi cations done with randomly chosen subsets. We note that classi cations done with well classied elements tend to fall closer to each other and algorithms tend to converge faster. We also observed some correspondence between the microbial nature of the data and the noise.