Machine Learning Classifiers applied to RKI MALDITOF Mass Spectrometry of Bacteria Data

Abstract

In this work we assess the performance of different machine learning methods, to classify bacteria from mass spectra available at RKI MALDI-TOF database. Microorganism identification employing mass spectrometry is a technology that has become very popular in the last years, especially in clinical microbiology, where an adequate classification is relevant to choose a proper treatment. The techniques selected in our study include discriminant analysis, decision trees, nearest neighbors and neuronal networks. We consider the following measures for the analysis: accuracy, Cohen κ coefficient, no information rate and time consumed. The obtained results allow us to recommend linear discriminant analysis, with a slightly lower performance than nearest neighbors, but with advantages in computational cost.