Outlier detection for machine olfaction based on odor-type signatures

A method for detecting or identifying odor outlier sample plays an essential role in implementing machine olfaction, called electronic nose (e-nose). The benefit of removing outlier not only eases the classification design process but also helps to improve the classification performance of the e-nose. In this study, odor-type signatures derived from the sensor array´s response waveforms are employed to detect the odor sample with high dimensionality that deviates in some degree from other odor samples. Four odor samples used for investigation consist of bacteria, coffee, soda, and rice with varying data quality. The experimental performance of the purposed method shows promising results to detect odor outlier.