A Machine Learning-Based Diagnostic Framework for Accurate Tuberculosis Detection

Abstract

Tuberculosis (TB) is a highly infectious disease with a profound impact on global health, making early and accurate diagnosis crucial for preventing its spread. In recent years, machine learning has emerged as a powerful diagnostic tool capable of enhancing disease detection and optimizing healthcare processes. This study presents a machine learning-based diagnostic model designed to improve the precision and efficiency of TB diagnosis. The proposed framework utilizes a comprehensive dataset comprising clinical and laboratory information, including patient demographics, symptoms, and diagnostic test results. The model operates through several key stages: data collection, preprocessing, feature extraction, feature fusion union, classification, and distribution. Experimental evaluation across multiple datasets demonstrates the model’s robustness and superior performance. On the TB-DS-I dataset, it achieved precision, recall, accuracy, and F-measure scores of 94.84%, 94.56%, 95.85%, and 94.32%, respectively. For the TB-DS-II dataset, it attained precision of 96.74%, recall of 96.5%, F-measure of 96.15%, and accuracy of 97.02%. Similarly, the TB-DS-III dataset yielded outstanding results with 97.83% precision, 97.54% recall, 97.2% F1 score, and 98.03% accuracy, confirming the model’s effectiveness and reliability in accurately classifying individuals as TB-positive or TB-negative.