Drug resistant tuberculosis classification using logistic regression

Tuberculosis (TB) is a communicable disease that poses a serious threat to the lives of people in the world. TB may be classified as drug-sensitive (DS-TB) or drug resistant (DR-TB). Clinical methods for diagnosing DR-TB require costly infrastructure as well as highly skilled expertise and also have major limitations due to the large amount of time taken to provide results. Researchers in the past have applied some machine learning methods in diagnosing TB, but not many studies have considered DR-TB and proved much satisfactory results. This paper examines the classification of DR-TB, based on existing symptoms, using logistic regression (LR) which is well known for its application in predictive diagnosis of diseases, and such approach is advantageous because they allow a quick evaluation and classification of disease outcomes in the presence of multiple and complex symptoms. To ensure that the symptoms selected have a direct relationship with DR-TB, a statistical test of hypothesis was established to prove the degree of association between the symptoms and DR-TB. In comparison to other techniques, the key advantages of the proposed LR classifier are its ability to show the degree to which a predictor variable affects the target variable and it works well with fewer feature variables. The comparative results with other machine learning techniques shows that LR classifier resulted in more accuracy score than others.