Data Reconstruction Using Subspace Analysis for Gas Classification

We propose a data reconstruction method using subspace analysis for gas classification in an electronic nose system. A noise generated by defects in sensors or by environmental factors in the process of data sampling significantly degrades data classification performance. In this paper, an electronic nose system more robust to data errors is designed by reconstructing lost or damaged values of data samples based on a statistical learning method that exploits the principal components. Diverse types of volatile organic compounds were employed in the classification experiments, which were conducted to reconstruct the lost or damaged data through the proposed method. The applied method prevented the degradation of classification performance and enhanced the discriminative power in the PCA+LDA feature space.