An Imbalanced Learning based MDR-TB Early Warning System

As a man-made disease, multidrug-resistant tuberculosis (MDR-TB) is mainly caused by improper treatment programs and poor patient supervision, most of which could be prevented. According to the daily treatment and inspection records of tuberculosis (TB) cases, this study focuses on establishing a warning system which could early evaluate the risk of TB patients converting to MDR-TB using machine learning methods. Different imbalanced sampling strategies and classification methods were compared due to the disparity between the number of TB cases and MDR-TB cases in historical data. The final results show that the relative optimal predictions results can be obtained by adopting CART-USBagg classification model in the first 90 days of half of a standardized treatment process.