Tuberculosis (TB) remains one of the most fatal infectious diseases, particularly in developing countries where healthcare accessibility is inadequate. Although conventional diagnosis methods like sputum smear microscopy and chest radiographs are still prevalent, they lack sensitivity, are subjective, and remain difficult to access in rural communities. Medical imaging based on deep learning has also been a strong remedy to automate TB diagnosis with enhanced accuracy and efficiency. Yet, image variation and class imbalance are still significant issues. In this paper, we introduce a thorough framework consisting of data augmentation and supervised ML models to classify the chest X-ray images into Normal and TB-positive categories. With a curated and enriched dataset of 7,000 images, these models were tested for accuracy and generalization, with fine KNN and Naïve Bayes producing the best accuracy of 94.44%. These findings indicate that traditional machine learning models, when appropriately tuned and balanced, can be used as lightweight yet dependable tools for TB detection, especially in low-resource settings.