Colorimetric sensor array optimization using cluster analysis

Colorimetric sensor arrays have proven to be simple yet efficient detection tools for on-site diagnostics of wide range of gases. Nonetheless, using a large number of sensors does not guarantee a better recognition for the specified gases but on the contrary some sensors responses are extremely deviated with noise and are either irrelevant or redundant to the recognition process. For a better accuracy, these sensors should be recognized and removed from the sensor array. This paper addresses the selection of the optimal sensors set based on the selectivity differences of each sensor by dividing the array into subarrays using hierarchical cluster analysis (HCA) and showing the discrimination accuracy using principal component analysis (PCA). The proposed methodology is applied to a colorimetric sensor array dataset and proves to significantly reduce the number of sensors while maintaining an excellent recognition accuracy.