In this study, antimicrobial resistance of P. aeruginosa was predicted using a data mining assessment framework by machine learning algorithms, as shown in Figure 1 . There were a total of six stages involved in reaching these conclusions, including the following: objective; data collection and preparation; machine learning techniques on a data mining platform; model building; evaluation and assessment; and implications. Initially, we collected the data and did some preliminary preprocessing to pick the right attributes. Afterward, this data was used for analysis and assessment. Secondly, Weka (v3.9.2), “a java-based machine learning and data mining platform,” was used to measure and evaluate classifications with the most recent bio-Weka and RF plugins. In addition, the results of machine learning classifiers were used in logistic regression (LR) to evaluate the resistance phenotype assessment to twelve different antibiotic drugs, namely, ampicillin, amoxicillin, meropenem, cefepime, fosfomycin, ceftazidime, chloramphenicol, erythromycin, tetracycline, gentamycin, butirosin, and ciprofloxacin.
Furthermore, the data was divided into two sets (training set and testing set) by a ratio of 60 : 40. Overfitting was prevented by using 10-fold cross-validation, and training data were used further as efficiently as possible to determine the optimal hyperparameter settings. The training model's evaluation results were based on an average of the hyperparameter values that fared best in the 10-fold scross-validation procedure. Sensitivity, specificity, accuracy, and precision were used to assess the model performance of bio-Weka and RF by equations (1 )–(4 ). The number of strains that turned out to be resistant was the true positive (TP), the number of strains that turned out to be sensitive was the true negative (TN), and the number of strains that turned out to be resistant when they should have been sensitive was the false positive (FP), and the number of strains that should have been sensitive when they should have been resistant was the false negative (FN) [36 (link)].
Furthermore, the data was divided into two sets (training set and testing set) by a ratio of 60 : 40. Overfitting was prevented by using 10-fold cross-validation, and training data were used further as efficiently as possible to determine the optimal hyperparameter settings. The training model's evaluation results were based on an average of the hyperparameter values that fared best in the 10-fold scross-validation procedure. Sensitivity, specificity, accuracy, and precision were used to assess the model performance of bio-Weka and RF by equations (
Free full text: Click here
