A novel electronic nose learning technique based on active learning: EQBC-RBFNN

Electronic nose (E-nose) is an intelligent system that we will use in this paper to distinguish different kinds of indoor pollutant gases. We usually need to obtain a large number of the labeled samples, and then E-nose can learn enough useful information from them which can help it make a right decision. In fact, it is always a time-consuming and laborious thing to label the unlabeled samples, and the numbers of collected unlabeled samples are often far greater than that of the labeled samples. On the other hand, if we only use a small number of labeled examples and ignore the valuable information included in the unlabeled samples, the E-nose learning system usually does not have a strong generalization. So an active learning (AL) algorithm based on improved query by committee (QBC) for RBFNN is proposed and applied to E-nose, which is called as EQBC-RBFNN. This technique efficiently combines QBC and RBFNN for E-nose training with both labeled and valuable unlabeled samples. And we employ this method to train the E-nose which is used to distinguish three indoor pollutant gases (toluene, formaldehyde and benzene). The results of data processing prove that in this way, the classification accuracy of the E-nose has been improved when unlabeled samples in application process are used to refine the E-nose compared to an E-nose trained only with labeled samples.