Unsupervised gas discrimination in uncontrolled environments by exploiting density peaks

Gas discrimination with Open Sampling Systems based on low-cost electro-chemical sensor arrays is of great interest in several applications, such as exploration of hazardous areas and environmental monitoring. Due to the lack of labeled training data or the high costs of obtaining them, and the presence of unknown interferents in the target environments, supervised learning is often not applicable and thus, unsupervised learning is an attractive alternative. In this work, we present a cluster analysis approach that can infer the number of different chemical compounds and label the measurements in a given uncontrolled environment without relying on previously acquired training data. Our approach is validated with data collected in indoor and outdoor environments by a mobile robot equipped with an array of metal oxide sensors. The results show that high classification accuracy can be achieved with a rather low sensitivity to the selection of the only functional parameter of our proposed algorithm.