An optimizing method based on Water-Filling for Case Attribute weight

Case retrieve is the key link in Case-Based Reasoning (CBR) system, and the distribution of attribute weights affects the retrieval accuracy directly. However, the traditional retrieval methods do not pay much attention to the weights distribution which led to low retrieval efficiency. In this paper, a new method based on Water-Filling principle in wireless communication field is proposed to optimize the case attribute weights. Regarding every single attribute in the case base as a sub channel, then calculates the importance of each case attribute by analyzing the data volatility with the Water-Filling principle. To prove the availability of the method, a glass identification dataset from the UCI database is used for a simulation experiment and the result illustrates that the new method could get a better and more accurate retrieval result compared with the traditional methods. The method could dig the inner information of each case attribute, and could assign proper weight to each attribute, which improves the retrieval accuracy and proves that the method the paper introduced is effective for CBR system.