Signal drift caused by sensors or environmental changes, which can be regarded as data\ndistribution changes over time, is related to transductive transfer learning, and the data in the target\ndomain is not labeled. We propose a method that learns a subspace with maximum independence of\nthe concentration features (MICF) according to the Hilbert-Schmidt Independence Criterion (HSIC),\nwhich reduces the inter-concentration discrepancy of distributions. Then, we use Iterative Fisher\nLinear Discriminant (IFLD) to extract the signal features by reducing the divergence within classes\nand increasing the divergence among classes, which helps to prevent inconsistent ratios of different\ntypes of samples among the domains. The effectiveness of MICF and IFLD was verified by three\nsets of experiments using sensors in real world conditions, along with experiments conducted in the\nauthorsâ?? laboratory. The proposed method achieved an accuracy of 76.17%, which was better than\nany of the existing methods that publish their data on a publicly available dataset (the Gas Sensor\nDrift Dataset). It was found that the MICF-IFLD was simple and effective, reduced interferences,\nand deftly managed tasks of transfer classification.
Loading....