A Weighted Classification Method Based on Adaptive Feature Selection

This paper proposes a new classification method, which can adaptively select effective features and give different classification weights according to the classification ability of features, this method effectively eliminates the influence of invalid features in classification process. For the classification problem of fault diagnosis, this method judges the fault type from the azimuth relative to the normal state in multi-dimensional space, and judges the fault degree from the ratio of distance, the method can not only give the fault category, but also give the fault probability, fault degree and whether there is a new fault type. In the process of updating the classification center, the classification method proposed in this paper solves the problem that the location of the classification center changes with the sample processing order. The effectiveness of this method is tested by the transformer vibration data (TVIB) and five data sets published in MATLAB and UCI, and compared with K-Mean, self-organizing map (SOM) methods, it is proved that considering the classification weight can effectively improve the classification accuracy, and the proposed method has strong applicability for different classification problems and good processing ability for large capacity and multi feature data.