Utilization of Machine Learning Techniques for the Identification of Enterobacteriaceae in Urinary Tract Infections (UTI) Using Antibiotic Susceptibility Testing Data

This paper aims to compare the best machine-learning algorithm for the classification of three Enterobacteriaceae genera, namely Proteus, Klebsiella, and Escherichia from antibiogram data. Random forest classification, XGboost, and artificial neural network algorithms were developed for classification purposes and trained on the dataset acquired from Clinical University Hospital Mostar in the period from 2017 to 2022 comprising 6690 distinct entries. Specifically, we trained models on two sets of data: the original, unprocessed dataset, and a dataset subjected to random subsampling. Our ANN model has demonstrated robust performance on unprocessed datasets, achieving an accuracy rate of 81.2579% when distinguishing between three bacterial genera within the Enterobacteriaceae family. Additionally, the ANN model achieved an accuracy rate of 93.81% in classifying Escherichia, outperforming the random forest model, which attained an accuracy of 93.07%. However, the random forest model exhibited superior robustness across other bacterial genera. Additionally, eXplainable Artificial Intelligence (XAI) techniques, particularly SHAP (Shapley Additive exPlanations), were employed to interpret model decisions. The application of SHAP revealed insightful findings: susceptibility to F/M characterized Escherichia Coli, whereas Proteus and Klebsiella mostly exhibited resistance to the F/M. By harnessing ANN methods we can significantly improve our ability to identify bacterial genera, possibly leading to more effective diagnostic and treatment strategies in microbiology. This research lays the groundwork for future investigations that may encompass larger datasets and aim to distinguish among a wider array of bacterial genera.