A Prototype Sensor-Based System with Machine Learning for Cannabis Strain Classification via VOC Signatures

Cannabis strain classification is essential for ensuring product consistency, therapeutic accuracy, and quality control in both medical and commercial applications. This study presents the development of a prototype sensor-based system for classifying three cannabis strains—Black Patronas (Hybrid), MAC Gold (Indica), and Banana Daddy R1 (Sativa)—by analyzing odors emitted from dried flowers and leaves. A gas sensor array consisting of five low-cost sensors (MQ-2, MQ-3, MQ-6, TGS-822, and TGS-826) was employed to detect volatile organic compounds (VOCs) characteristic of each strain. Sensor signals were acquired using an Arduino Mega 2560, preprocessed via Node-RED, stored in InfluxDB, and visualized using Grafana. To enable classification, differential responses (ΔR) were computed by subtracting baseline analog values from VOC exposures. These ΔR values were used to train a Random Forest classifier, which achieved an accuracy of 83% on unseen test samples. Notably, MQ-2 showed strong response to Hybrid flowers, while TGS-822 was most effective for detecting VOCs from Sativa leaves. While the dataset was limited in size, the system demonstrated reliable classification across six cannabis sample types. These results confirm the feasibility of using low-cost sensor arrays with interpretable ML models for odor-based strain identification. Future work will expand the dataset, explore additional algorithms, and move toward real-time, portable deployment for cannabis quality assurance.