Noise-Robust Hybrid Machine Learning and Deep Learning Framework for Early Tuberculosis Detection from Cough Audio Recordings

Abstract

Tuberculosis (TB) remains a major global health challenge, particularly in regions with limited access to rapid and reliable diagnostic facilities. Traditional diagnostic methods are often time-consuming, expensive, and require specialized infrastructure, which delays early detection. To address these limitations, this paper proposes a noise-robust hybrid framework for early tuberculosis detection using cough audio recordings. The system integrates preprocessing techniques for noise reduction, feature extraction using Mel-Frequency Cepstral Coefficients (MFCCs) and spectrograms, and classification using both machine learning models (Support Vector Machine, Random Forest) and deep learning architectures (Convolutional Neural Networks and CNN–RNN). An ensemble decision-making approach is employed to combine predictions from multiple models and improve reliability. Experimental results demonstrate that the proposed method achieves an accuracy of 94.8% and performs effectively even in noisy real-world conditions. The framework provides a scalable, cost-effective, and non-invasive solution for tuberculosis screening, making it suitable for deployment in resource-constrained environments.