Identification and Prediction of Type of TB Based on Drug Resistance Using Machine Learning

Tuberculosis (TB) has remained a major health concern and is the second deadliest infectious disease worldwide after Covid-19. This situation demands innovative approaches for detection and treatment, particularly in developing countries with limited resources for confirmatory tests. Traditional methods of diagnosis of the disease are based on multiple types of samples and various type of tests for each sample. In this study, we try to identify presence/absence of TB in an individual based on microbiological test results such as Microscopy, Culture and DST from the NIAID dataset which consist of patient cases from eleven countries of Eastern Europe, Asia and sub-Saharan Africa. Once presence of TB is confirmed, we try to classify the type of TB such as Mono, Poly, Multi or Extensive-Drug Resistance. Diagnosing of type of TB determines the drugs to be administered which is unique for each suspected individual. The research methodology involved data collection, preprocessing, and training ML models like Random Forest, Logistic Regression, Gradient Boosting classifier and KNN to predict TB types and identify drug resistance. The proposed method can be employed for identification of type of TB and suggest medication regimen based on DST results.