This study evaluates whether machine learning (ML) models can accurately predict early treatment outcomes in patients with multidrug-resistant or rifampicin-resistant tuberculosis (MDR/RR-TB), and aims to improve the ability of ML models in assessing early therapeutic efficacy. Zhang et al. conducted a retrospective study using demographic and clinical data from an internal cohort of 744 MDR/RR‑TB patients and an external validation cohort of 137 patients from hospitals in Beijing and Guangzhou, China. They developed logistic regression and seven machine learning models to predict sputum culture conversion at 2 and 6 months, and compared each model’s performance using AUC, accuracy, sensitivity, and specificity. As a result, culture conversion rates were high, reflecting strong early treatment responses. 81.9% of patients achieved conversion at 2 months and 87.1% at 6 months. Among the models tested, the artificial neural network (ANN) demonstrated the best predictive performance, with AUCs of 0.82 and 0.90 and accuracy, sensitivity, and specificity exceeding 0.74 at 2 and 6 months, respectively. Some of the k factors that influenced early conversion included mediastinal lymphadenopathy, medication compliance, sputum smear grading, and treatment regimens containing BDQ or LZD. These findings highlight the ANN model’s potential in aiding with efficient clinical decision-making and optimizing MDR/RR‑TB therapy. Overall, ML models consistently outperformed traditional logistic regression in both stability and generalizability. These ML models could serve as tools for early evaluation of therapeutic efficacy in patients with MDR/RR-TB, enabling clinicians to identify high-risk individuals, tailor treatment plans, and improve overall recovery rates.