Feature analysis and prediction of antibiotic resistance using PCA-MLP with segments of N. gonorrhoeae bacteria's DNA

With the explosive growth of biomedical data, several smart automated systems have evolved to discover knowledge and aid decision makers to improve quality of life of individuals and societies. In this paper, we concentrate on the prediction of antibiotic resistance using segments of Neisseria gonorrhoeae bacteria's DNA sequences. The aim is to build more effective prediction systems by using feature analysis to identify an optimal set of factors to reduce the complexity of predictive models. To assess the prediction effectiveness, the model is evaluated on two antibiotic datasets, one with imbalanced classes and another with very high dimensionality that significantly exceeds the number of available samples. The results demonstrate that dimensionality reduction is case-dependent and can range from 12% to 88% without significantly affecting the model accuracy.