Abstract

The Scent of Health (S-OH) dataset is the largest publicly available collection of clinical electronic nose (eNose) data for non-invasive disease screening via exhaled breath analysis. It comprises 1,234 patients across nine diagnostic groups (healthy controls and eight diseases: hepatitis, gastritis, fatty liver disease, diabetes, chronic renal failure, COPD, lung cancer, and tuberculosis), each providing a 17-channel multivariate time series of approximately 895 seconds recorded at 0.4 Hz. The dataset includes explicit temporal annotations over 13 consecutive weeks and two clinical sites, enabling reproducible research on sensor drift and cross-site generalization. All data are anonymized and provided in an AI-ready format with predefined temporal train/test splits. Baseline validation includes a 3-layer LSTM classifier for lung cancer screening, achieving AUC 0.691 under temporal evaluation, alongside additional CNN-based methods. The dataset and code are released under the MIT License and is publicly available at https://doi.org/10.57967/hf/9752. S-OH is designed to support a range of machine learning tasks, including binary and multi-class time series classification, drift-robust learning, and demographic bias analysis for breath-based diagnostics.