Artificial Breath Classification Using XGBoost Algorithm for Diabetes Detection

• Published in: Sensors (Basel, Switzerland) Vol. 21; no. 12; p. 4187
• Main Authors: Paleczek, Anna, Grochala, Dominik, Rydosz, Artur
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 18.06.2021; MDPI
• Subjects: Algorithms; Biomarkers; breath acetone; Carbon dioxide; Diabetes; Disease; Gases; Humidity; Machine learning; Mass spectrometry; Metabolism; Neural networks; Patients; Principal components analysis; Scientific imaging; Sensors; VOCs; XGBoost; United States > US; Switzerland; China; Japan
• ISSN: 1424-8220; 1424-8220
• DOI: 10.3390/s21124187

Exhaled breath analysis has become more and more popular as a supplementary tool for medical diagnosis. However, the number of variables that have to be taken into account forces researchers to develop novel algorithms for proper data interpretation. This paper presents a system for analyzing exhaled air with the use of various sensors. Breath simulations with acetone as a diabetes biomarker were performed using the proposed e-nose system. The XGBoost algorithm for diabetes detection based on artificial breath analysis is presented. The results have shown that the designed system based on the XGBoost algorithm is highly selective for acetone, even at low concentrations. Moreover, in comparison with other commonly used algorithms, it was shown that XGBoost exhibits the highest performance and recall.