Impact of sensing film's production method on classification accuracy by electronic nose

Abstract

The development of gas sensing materials is relevant in the field of non-invasive biodevices. In this work, we used an electronic nose (E-nose) developed by our research group, which possess versatile and unique sensing materials. These are gels that can be spread over the substrate by Film Coating or Spin Coating. This study aims to evaluate the influence of the sensing film spreading method selected on the classification capabilities of the E-nose. The methodology followed consisted of performing an experiment where the E-nose was exposed to 13 different pure volatile organic compounds. The sensor array had two sensing films produced by Film Coating, and other two produced by Spin Coating. After data collection, a set of features was extracted from the original signal curves, and the best were selected by Recursive Feature Elimination. Then, the classification performance of Multinomial Logistic regression, Decision Tree, and Naïve Bayes was evaluated. The results showed that both spreading methods for sensing films production are adequate since the estimated error of classification was inferior to 4 % for all the classification tools applied.