This study explores the noninvasive detection of diabetes mellitus through exhaled human breath using an electronic nose (e-nose) integrated with Tiny Machine Learning (TinyML). The e-nose, equipped with Metal Oxide Semiconductor (MOS) sensors, was designed to analyze volatile organic compounds (VOCs) in exhaled breath, particularly acetone, a key biomarker for diabetes. The research involved 44 participants, including 22 healthy individuals and 22 diagnosed with various types of diabetes. The study evaluated three machine learning algorithms: XGBoost, Deep Neural Networks (DNN), and 1D-CNN. The XGBoost algorithm achieved 95% detection accuracy, while DNN and 1D-CNN yielded 94.44% accuracy. These results highlight the potential of combining e-noses with TinyML for real-time diabetes detection, offering a noninvasive and rapid diagnostic approach. However, the study also discusses limitations, such as sensor sensitivity to high relative humidity (RH) and the need for continuous calibration. Future research should focus on expanding the dataset, improving sensor reliability, and enhancing the interpretability of the models.This study explores the noninvasive detection of diabetes mellitus through exhaled human breath using an electronic nose (e-nose) integrated with Tiny Machine Learning (TinyML). The e-nose, equipped with Metal Oxide Semiconductor (MOS) sensors, was designed to analyze volatile organic compounds (VOCs) in exhaled breath, particularly acetone, a key biomarker for diabetes. The research involved 44 participants, including 22 healthy individuals and 22 diagnosed with various types of diabetes. The study evaluated three machine learning algorithms: XGBoost, Deep Neural Networks (DNN), and 1D-CNN. The XGBoost algorithm achieved 95% detection accuracy, while DNN and 1D-CNN yielded 94.44% accuracy. These results highlight the potential of combining e-noses with TinyML for real-time diabetes detection, offering a noninvasive and rapid diagnostic approach. However, the study also discusses limitations, such as sensor sensitivity to high relative humidity (RH) and the need for continuous calibration. Future research should focus on expanding the dataset, improving sensor reliability, and enhancing the interpretability of the models.