This study presents a robust method for automatically classifying tuberculosis (TB) in chest X-rays using advanced deep learning techniques, particularly neural networks. Given the inherent imbalances in medical datasets, we employ the Synthetic Minority Over-sampling Technique (SMOTE) to enhance training data, ensuring a more balanced learning experience. Our model is trained on a diverse dataset of chest X-ray images, specifically addressing class imbalances in TB detection. The neural network architecture is tailored to capture intricate TB manifestations in X-ray images. Through iterative training, the model learns to identify subtle TB-related features, achieving remarkable accuracy. Experimental results validate the effectiveness of our approach, even in the presence of imbalanced classes, highlighting the importance of SMOTE in improving model performance. This study contributes to leveraging advanced machine learning for early disease detection, particularly in resource-constrained settings. Our proposed model holds promise as a valuable tool for healthcare professionals, facilitating timely and accurate TB diagnosis from chest X-ray images. Integrating such intelligent systems into clinical workflows could enhance TB screening programs, leading to improved patient outcomes and a more effective public health response.