A Data Mining Procedure for Polymer Selection for Making Surface Acoustic Wave Sensor Array

The paper presents a data mining procedure for selection of polymer coatings for making surface acoustic wave (SAW) sensor arrays for electronic nose applications. The procedure is based on linear-solvation-energy-relationship (LSER) parameters for target vapors, interferents and polymers. The procedure combines selections from three independent selection methods: heuristic-that selects polymers according to the maximum values of LSER parameters, principal component analysis (PCA)-that generates clusters of polymers according to similarity of their chemical affinities towards target vapors/interferents and selects those with largest dissimilarity, and fuzzy c-means (FCM) clustering-that generates fuzzy clustering of polymers according to a fuzzy measure of similarity and selects those that define cluster centers. The principal component and fuzzy selections are based on treating the polymers as objects represented by vapors as variables in the LSER generated partition coefficient data. The selection criterion is that the polymers that appear in at least two subset selections made by the above mentioned three methods are the winners. A second round of selection can be made by repeating the same procedure after eliminating the first round selections from the original set. The quality assessment of subset selections in each round is made by simulating and analyzing sensor array responses for specific target applications. The procedure has been validated by analyzing 4 target identification tasks-explosives, chemical weapon agents, narcotics and drugs of abuse. The vapor pool is comprised of 15 target and 23 interferent volatile organic compounds of varied origins including building materials, vehicular emissions, soil, vegetation, body odor, plastics, paints etc. The efficacy of the procedure is demonstrated by simulating 3 prototype detection targets: bombs in public place, nerve gas use by fugitives, and narcotics trafficking.