Multiclass classification of n-butanol concentrations with k-nearest neighbor algorithm and support vector machine in an electronic nose

An electronic nose (e-nose) is a machine used for sensing and recognizing odors by using chemical sensors. The performance of e-nose depends on choosing correct sensor and correct pattern recognition algorithm according to application fields and kinds of the odors. In this study, different n-butanol concentrations sensed by 12 metal oxide gas sensors are classified by using multiclass support vector machine methods (SVM) and k-nearest neighbor (k-NN) algorithm. Focus in this paper is that the performances of these algorithms are increased with a decision tree structure. Therefore the proposed decision tree structure is applied to the electronic nose data for sensor subset selection and classification of the n-butanol concentrations. SVM and k-NN algorithms are tested for classification of different concentrations in this decision tree structure and ordinary structure. In addition to these, cross-validation technique is used for both increasing success of classification algorithms and assessing the results objectively. This study shows that the success of classification algorithms increase from 87% to 93% and 86% to 96% by using data of two sensors selected with the proposed decision tree structure for the k-NN and the SVM methods, respectively.