A comparative analysis of predictive models to assist in the early diagnosis of tuberculosis

Tuberculosis (TB), an airborne infectious disease, is a major factor that creates a double disease burden and complex medical and economic problems in Indonesia. To eliminate TB by 2030, Indonesia has developed a National Strategic Plan with strategies to optimize efforts for early detection and utilize research findings and technologies for the screening and diagnosis of TB. In accordance with the National Strategic Plan, this study aimed to develop an effective prediction model to assist in the early diagnosis of TB based on patient symptoms and laboratory test results. This study followed the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology and applied four classification algorithms: Decision Tree, Naïve Bayes, Random Forest, and Support Vector Machine (SVM), on a dataset constructed from TB patients at a hospital in Surabaya between 2020 and 2023. The best model was evaluated based on the recall, specificity, and geometric mean (G-mean) values. Based on comparative analysis, the Random Forest model generated the best performance with a G-mean value of 0.984 and classification accuracy of 98.6%, followed by the SVM model with a G-mean value of 0.984 and classification accuracy of 98.6%. These results suggest that the predictive model is beneficial for accelerating early diagnosis of TB, allowing medical professionals to provide immediate medical services and further treatment. This contribution is in line with the Sustainable Development Goals (SDGs) of Good Health and Well-being to end communicable diseases, including Tuberculosis, by 2030.