Optimization of Breath VOC Sensors Using Porous Anti-Humidity Membranes and Machine Learning-Based Humidity Prediction

Breath-sensing technologies offer a non-invasive approach for detecting biomarkers in exhaled breath, however, high moisture content in exhaled air often compromises sensor–analyte interactions, reducing sensitivity, selectivity, and reliability. In this study, porous polydimethylsiloxane (PDMS) and UiO-66@PDMS anti-humidity membranes were fabricated via a sugar-cube templating method and integrated with a commercial ethanol-sensitive Figaro gas sensor to systematically enhance sensor performance under high-humidity conditions. The sensors were tested with ethanol, acetone, and their mixture under dry (20–30% RH) and humid (70–80% RH, up to >90% RH) conditions. While porous PDMS membranes mitigated humidity effects, they significantly reduced analyte sensitivity. In contrast, UiO-66@PDMS membranes maintained high sensitivity to ethanol, achieving a 52% response to 50 ppm ethanol under humid conditions, outperforming the membrane-free sensor (41% response) and demonstrating enhanced selectivity and stability. To further improve sensor reliability, five supervised regression models—Gaussian process, random forest (RF), RF with hyperparameter optimization, k-nearest neighbors (KNN), and support vector machine (SVM) with hyperparameter optimization—were trained to predict relative humidity from sensor outputs. The RF model with hyperparameter optimization achieved 98% accuracy and R² = 0.998, providing robust correction of humidity-induced interference. Pearson correlation analysis showed a reduction from 0.708 to 0.190 after membrane integration, quantitatively confirming effective humidity mitigation. These results demonstrate that combining MOF-based anti-humidity membranes with machine learning enables reliable, accurate, and reproducible VOC detection in humid environments, offering a practical and scalable strategy for next-generation breath-based sensors.