Predicting the Smear Conversion of Pulmonary Tuberculosis Patients Using Machine Learning

This paper presents the development and evaluation of a neural network model for predicting the after-treatment smear status in a medical context. The dataset comprised demographic variables, X-ray image results, and the smear status after the intensive phase. To address the data imbalance, the SMOTE technique was employed. Significant variables were selected through correlation analysis and feature selection, leading to the construction of a neural network model with specific parameters. The model was trained on a 70% training dataset using 3-fold cross-validation and achieved an accuracy of 90.80% after 400 epochs. Model evaluation and validation on the remaining 30% of the testing dataset demonstrated its effectiveness in accurately classifying positive and negative smear statuses. The precision, recall, and F1-scores for both classes indicated balanced performance. The model's reliability in predicting smear status is supported by the heatmap analysis. Future work includes further validation on larger and more diverse datasets, consideration of potential limitations, and exploration of advanced techniques like deep learning. The developed neural network model shows promise in clinical applications and can contribute to improved decision-making in the treatment of patients.