Boosting-based one-class SVM for recognizing true-fake Chinese liquors using electronic noses

The true-fake detection of Chinese liquors can be considered as a one-class classification problem. Using only positive sample in training process brings about difficulties in determining the optimal parameters of one-class SVM and its kernel. And one-class SVM optimized with conventional grid search method has a low recognition rate of positive test samples. Hence, we proposed a new idea for recognition, namely boosting-based one-class SVM. Six different kinds of liquors were sampled by a self-designed electronic nose system. After the preprocessing and feature extraction of sample data, 18 groups of optimum parameters of one-class SVM were chosen from specified parameter range, each group was used to train a classifier with training algorithm. Then the true-fake decisions were made for test samples with boosting integration rules. Finally the recognition rates of Hongjinjiu and Lanjinjiu positive sample reached 95.24% and 100%, and the recognition rates of both positive and negative sample reached 97.04% and 94.83%, respectively.