Classification of multiple indoor air contaminants by an electronic nose and a hybrid support vector machine

This paper presents a laboratory study of multi-class classification problem for multiple indoor air contaminants which belongs to a completely linear-inseparable case. Six kinds of indoor air contaminations (formaldehyde, benzene, toluene, carbon monoxide, ammonia and nitrogen dioxide) were recognized as indicators of air quality in this project. The effectiveness of the proposed HSVM model has been rigorously evaluated on the experimental E-nose data sets. In addition, we have also compared it with existing five methods including Euclidean distance to centroids (EDC), simplified fuzzy ARTMAP network (SFAM), multilayer perceptron neural network (MLP) based on back-propagation learning rule, individual FLDA and single SVM. Experimental results have demonstrated that the HSVM model outperforms other classifiers in general. Also, HSVM classifier preliminarily shows its superiority in solution to discrimination in various electronic nose applications.