Machine learning-based artificial nose on a low-cost IoT-hardware

Abstract In order to make Internet of things applications easily available and cost-effective, we aim at using low-cost hardware for typical measurement tasks, and in return putting more effort into the signal processing and data analysis. By the example of beverage recognition with a low-cost temperature-modulated gas sensor, we demonstrate the benefits of processing techniques in big data such as feature selection and dimensionality reduction. Specifically, we determine a subset of temperatures that yields good support vector machine classification results and thereby shortens the measurement process.