Machine Learning Modelling of Antibiotic Resistance of Faeca Microbial Isolates of Muturu and White Fulani Cattles

Abstract

Antimicrobial Resistance (AMR) is already a global health concern and challenging pharmaceutical progress in the development and design of antibiotics. It has also been reported by WHO to be a major threat and thus the need to find a solution to this public health issue through modelling of this resistance. This research thus investigated the Machine Learning (ML) antimicrobial modelling of antimicrobial resistance characterization of the faecal samples of muturu and white Fulani cattle. This experiment involved collecting 40 cattle faecal microbial isolates. The isolated microbes (40 isolates) were subjected to antimicrobial resistance tests, and the results were modelled through machine learning algorithms. The results showed that this model was dependent on specific antibiotics. Some models, such as Naïve Bayes, performed excellently for perfloxacin, Augmentin, with accuracy of 1.00±0.00 and Random Forest (RF) for Chloramphenicol, Amoxicillin, and Gen-tamcyin with high accuracy (0.92±0.08). While ensemble tree methods (RF, decision tree, and gradient boosting) achieve 100% accuracy, Support Vector Classifier (SVC) remained the most resilient generalized predic-tor. For the bacteria isolates from white Fulani, only support vector classifier was consistent in predicting AMR across all the tested antibiotic. The high standard deviation observed in some drugs like oflaxocin and cotrimethazol could be due to the dataset or the microbial interaction with drugs. There is therefore a need to build reliable ML-driven AMR pipelines that should address critical global heath burden. Poorly managed Fulani cattle herding risks antibiotic abuse, but accurate modeling provides vital decision support.

DOI: https://doi.org/10.5281/zenodo.22916540