Complete Dataset and Source Code for Machine Learning-Based Identification of Factors Associated with False-Negative TSPOT.TB Results in Pulmonary Tuberculosis

The study enrolled 6,244 patients with pulmonary tuberculosis, of whom 810 were selected via random undersampling for model development. The dataset includes 63 clinical features (e.g., peripheral blood laboratory indices and demographic characteristics) and TSPOT.TB test results (true negative/false negative).

LASSO regression and the Boruta algorithm were employed for feature selection. Six machine learning models were constructed based on the selected features, and model performance was evaluated using accuracy, precision, recall, and F1 score. The SHAP technique was applied to interpret the optimal model.This dataset has been rigorously de-identified; all direct and indirect personal identifiers have been removed.For the runtime environment and dependencies, please refer to requirements.txt and README.md in the archived package.When using this dataset for research or publication, please cite the original paper.This study was performed in accordance with the Declaration of Helsinki.