A Mixed Similarity Measure for Data with Numeric, Symbolic and Ordinal Attributes(Artificial Intelligence I)(Joint Workshop of Vietnamese Society of AI, SIGKBS-JSAI, ICS-IPSJ, and IEICE-SIGAI on Active Mining)

概要

論文の詳細を見る
Many methods of knowledge discovery in databases are distance-based, such as instance-based learning or clustering where similarity measures between objects plays an essential role. Besides, it is known that most our real-word data not only contain numeric, symbolic, and ordinal attributes individually but also carry all of them in mixed way. Therefore, a Mixed Similarity Measure (MSM) for numeric and symbolic attributes is not enough for various data process. Moreover, the high cost of O (n^2logn^2) and O (n^2) for time and complexities of the existing algorithms do not allow the MSM to be applied to large datasets in KDD. As a result, we have proposed a fast algorithm to compute the Goodall's MSM for numeric and symbolic attributes in a linear complexity. In this paper, as an extension of the MSM, we consider an MSM^* for numeric, symbolic and ordinal attributes and describe a fast algorithm for MSM^* with a linear complexity as well. The experimental results show that the proposed MSM^* is also better than MSM and C4.5/See5.0 for the classification problem.
社団法人電子情報通信学会の論文
2004-11-27