Predicting Tuberculosis Treatment Outcomes using Data Mining Technology.
No Thumbnail Available
Date
2013-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa University
Abstract
Background: Tuberculosis is the second most common causes of death throughout the world next to HIV/AIDS. Ethiopia is also among the high burden countries. Though the disease has been a cause of death for millions of people around the globe, it is curable. Prediction of treatment outcome of TB patients using data mining techniques help the effort to stop TB-health problem.
Objective: The objective of this research was to prepare a predictive model for TB treatment outcomes that assist clinical decisions in connection with TB treatment.
Method: The six steps Ciso et al Hybrid Model were used. A total of 6332 instances were collected from five health centers of Addis Ababa City Government that provide tuberculosis treatment. A pre-processed the data was fed in to data mining tools with selected classification algorithms. These algorithms were J48, Naïve Bayes, SMO and PART. Accuracy and Area under ROC were the metrics used to compare models generated by the algorithms.
Result: After successive experiments using the four algorithms, PART algorithm revealed best performance. An accuracy of 81.32% and area under ROC=0.89. The algorithm generated five rules for the three treatment outcomes and the rules were found to be interesting for experts.
The rules contain the following predictor variables for treatment outcome: HIV Status, Sex, Age,Initial Weight with second month weight and Patient Category.
Conclusion: The findings from the research indicated that for the tuberculosis dataset with class imbalance PART found to be the best learner algorithm and most importantly clinical decisions such as diagnosis, prognosis and resource allocation can be supported by data mining techniques.
Description
Keywords
Tuberculosis,Data Mining