Analysis of Human Breath with E-Nose: Diabetes Mellitus Identification via SVD-XGBoost Algorithm

Abstract

Diabetes is a highly prevalent chronic disease worldwide. Analysis of human breath has emerged as a non-invasive method for detecting various conditions using biomarkers. Electronic noses represent a crucial tool in this breath analysis for patients with diabetes mellitus, enabling early detection and diagnosis. This study involves the evaluation of 22 healthy patients and 20 patients with diabetes mellitus using an electronic nose employing catalytic Metal-Oxide-Semiconductor (MOS) gas sensors. A computational algorithm based on Singular Value Decomposition (SVD) for feature extraction and selection was utilized, coupled with classification using the Extreme Gradient Boosting (XGBoost) algorithm. The results demonstrate that classification of singular value vectors with the XGBoost algorithm achieves an accuracy of 95.24identifying healthy and diabetic patients. This approach shows significant potential for early diagnosis of diabetes through breath analysis, highlighting the effectiveness of electronic nose technology alongside advanced computational techniques in distinguishing between patient groups.