PREDICTING TREATMENT UNFAVOURABLE IN PULMONARY TUBERCULOSIS PATIENTS USING STACKING ENSEMBLE MACHINE LEARNING APPROACH
The leading infectious disease-related cause of mortality for people is tuberculosis (TB). India is one of the countries with the highest rates of TB worldwide, making it a serious public health problem. People with active lung TB can spread the illness by spitting, coughing, or sneezing. In healt...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Institute of Mechanics of Continua and Mathematical Sciences
2025-05-01
|
| Series: | Journal of Mechanics of Continua and Mathematical Sciences |
| Subjects: | |
| Online Access: | https://jmcms.s3.amazonaws.com/wp-content/uploads/2025/05/14184914/jmcms-2505003-Predicting-Treatment-Unfavorable-Shaik.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The leading infectious disease-related cause of mortality for people is
tuberculosis (TB). India is one of the countries with the highest rates of TB
worldwide, making it a serious public health problem. People with active lung TB can spread the illness by spitting, coughing, or sneezing. In healthcare, the application of machine learning (ML) that helps in diagnosis is on the rise. In this study, we suggest a stacked ensemble model that combines three base ML classifier models to predict treatment-unfavorable in Pulmonary TB (PTB) patients. Cases with unfavorable treatment are considered as the event of interest. Retrospectively, secondary data of 1236 PTB patients treated in randomized controlled clinical research were obtained and split into training and testing data in a 70:30 ratio. Several ML models had different levels of effectiveness in predicting treatment-unfavorable outcomes in PTB patients. The Support Vector Machines model struggled with sensitivity (0.246) but had high specificity (0.981). Likewise, the Logistic Regression model showed poor
sensitivity (0.339) but strong specificity (0.959). The Decision Tree model, on the
other hand, did well, with high sensitivity (0.755) and specificity (0.956). With the
best accuracy (0.929), sensitivity (0.774), specificity (0.956), and F1-score (0.759),
the stacked Ensemble Random Forest model performed better than the others. This illustrates the prospective of ensemble learning in the healthcare industry, where it is essential to identify negative effects early and accurately. To improve prediction accuracy and generalizability, future research should verify these results and explore other clinical characteristics. |
|---|---|
| ISSN: | 0973-8975 2454-7190 |