GA-based optimal feature weight and parameter selection of NPPC for tea quality estimation

Electronic nose (e-nose) is an artificial olfaction system that is being widely used in many industries. E-noses detect smells with the help of electronic signals produced by a number of sensors. The important part of an efficient e-nose system is to recognize these electronic signals accurately by some pattern classification algorithm. Recently developed nonparallel plane proximal classifier (NPPC) has shown its effectiveness in pattern classification task using kernel trick. In general the performance of such classifier depends on the values of optimal parameter set as well as the feature set. In this research work we have studied the effect of simultaneous parameter and feature weight selection on the accuracy of black tea quality estimation employing multiclass one vs. one NPPC. In order to choose the model parameters we have used genetic algorithm (GA). Experimental results show that GA-based tuning and feature weighting scheme increases the performance of NPPC by ~ 2% in the problem of black tea quality prediction.