Intelligent Sensing Using Metal Oxide Semiconductor Based-on Support Vector Machine for Odor Classification

Abstract

Classifying odor in real experiment presents some challenges, especially the uncertainty of the odor concentration and dispersion that can lead to a difficulty in obtaining an accurate datasets. In this study, to enhance the accuracy, datasets arrangement based on MOS sensors parameters using SVM approach for odor classification is proposed. The sensors are tested to determine the sensors' time response, sensors' peak duration, sensors' sensitivity, and sensors' stability when applied to the various sources at different range. Three sources were used in experimental test, namely: ethanol, methanol, and acetone. The gas sensors characteristics are analyzed in open sampling method to see the sensors' performance in real situation. These performances are considered as the base of choosing the position in collecting the datasets. The sensors in dynamic experiment have average of precision of 93.8-97.0%, the accuracy 93.3-96.7%, and the recall 93.3-96.7%. This values indicates that the collected datasets can support the SVM in improving the intelligent sensing when conducting odor classification work.