An Efficient Machine Learning Techniques as Soft Diagnostic for Tuberculosis Classification Based on Clinical Data

Abstract

TB infection is a global problem, especially in Yemen. Early detection is critical to reducing TB deaths. As a result, accurate tuberculosis diagnosis takes time due to numerous clinical examinations. This problem requires a new diagnosis schema. In this study, we proposed classification models based on Efficient Machine Learning Techniques (EMLT), which predict whether the patient is TB-positive or TB-negative. Nine Different Efficient Machine learning models were trained and tested in two imbalance dataset cases using Stratified 10-Fold Cross-Validation and Holdout Cross-Validation and balanced dataset case using Holdout Cross-Validation with Synthetic Minority Oversampling Technique (SMOTE). The best model was evaluated on a test set using F1-score measure in imbalanced dataset case and accuracy measure in balanced dataset case. Based on the obtained results, the models that achieved the highest value of the F1-Score measure in the imbalanced dataset were LR and GBC with 99.826% value in Stratified Cross-Validation approach and GBC with 86.0334 in the Holdout Cross-Validation approach. And the models that achieved the highest value of the accuracy measure in the balanced dataset case (SMOTE) and Holdout Cross-Validation, were LR and GBC with a 99.725% value.