ML–SDR Framework for Selective and Rapid Detection of Volatile Organic Compounds

Volatile organic compounds (VOCs) pose significant environmental and health risks, yet their rapid and selective detection remains a persistent challenge due to the limited sensitivity and real-time adaptability of conventional sensors. This work presents a machine learning-integrated software-defined radio (ML–SDR) framework for the selective and rapid detection of VOCs using a 2.4 GHz coupled block resonator-based microwave sensor with a milled channel. The sensing element is interfaced with a custom-designed power amplifier (PA) and a Raspberry Pi–controlled SDR to achieve a compact and real-time measurement setup. Each VOC sample of 100 μL was introduced into the sensing channel, and its attenuation, group delay, and coupling factor responses were recorded. Distinct variations in these parameters were observed across eight VOCs, namely distilled water, acetone, chlorobenzene, chloroform, ethanol, isopropanol, methanol, and toluene, demonstrating the dielectric sensitivity of the proposed structure. A supervised learning model based on the Random Forest algorithm achieved a classification accuracy of 98.25%, with multi-class ROC curves exhibiting an average AUC exceeding 0.9. The calculated limit of detection (LOD) of the sensor was approximately 7.3 μL, confirming its high resolution and selectivity toward dielectric perturbations. This integrated ML–SDR system thus establishes a versatile and scalable platform for on-field VOC monitoring with high reliability, real-time response, and minimal hardware complexity.