2011 年 16 巻 4 号 p. 677-686
In this report we evaluate a portable system that records scent environment under daily life conditions. The system consists of an array of gas sensors, a fan for air flow control, a power supply and associated electronics for the sensors and the logging system. The system records intensity of environmental gas vapors. An air fan was set at the rear side of the sensor array to provide with a constant air flow for stable and reproducible detection. We evaluate the possibility of classification by methods of PCA and clustering on a recorded dataset of 21 different feature points found in daily life. Although the classification results projected on the two principal components do not achieve identification of the feature points, clusters of feature points form that are characteristic of the feature point environmental conditions such as environments related to foods, outdoor, natural outdoor and indoor. Additionally with a view to reduce the system size and components, we estimate small combination of sensors by numerical investigation of variable selection in PCA.