Integrated sensor array optimization with statistical evaluation

In a sensor array system, sensor selection is to select optimum sensor elements which as a whole can maximize the separability of sensor array responses to different analytes. The purpose is to improve the quality of input to a classifier, thereby, improve correct classification rate. This paper introduces an integrated approach of cluster analysis (CA) and genetic algorithms (GA) to find the optimal subset of sensors in the array which can provide maximum diversity in response to certain analytes, on the premise that larger diversity of sensors leads to better classification performance. The cluster analysis is used to identify the number of sensors to form an array. The results obtained from the cluster analysis are then used by the genetic algorithm to obtain the best sensor subset. A statistical criterion, general resolution factor (GRF), is proposed in this study to evaluate the optimization results in the feature spaces. That statistical measure can quantitively evaluate the quality of input space. Moreover, the quality for input spaces of different dimensions can be compared based on GRF with the aid of principal component analysis. The results indicate that the proposed optimization procedure can successfully identify a sensor subset providing improved input for classification.