基于混合属性模糊相似度的k-近邻分类器

k-NN Classifier Based on Hybrid-Attribute Fuzzy Similarity

摘要: 在医学诊断等实际应用中，广泛存在着数值型、区间型与分类型属性共存的混合数据分类问题。现有方法往往难以充分融合与利用此类异构数据的原始信息，导致分类器性能不佳，无法满足实际应用中对精度与稳健性的要求。为此，文章提出一种基于混合属性模糊相似度的分类器。首先，针对混合数据结构，构建适配属性的模糊相似度；进而基于乘积t-范数，建立一种能够统一处理多类属性的模糊相似度度量。其次，在此基础上设计混合属性模糊相似度分类器，以更有效地利用数据的内在结构与语义信息进行分类。最后，为验证所提分类方法的有效性，将其与最大正区域分类器、线性支持向量机、多层感知机等5种代表性分类器进行对比实验。结果验证了新方法在多个数据集上的优越性能，为混合数据分类问题提供了一种有效的新途径。

Abstract: In practical applications such as medical diagnosis, there exists a widespread problem of classifying mixed data with coexisting numerical, interval, and categorical attributes. Existing methods often fail to fully fuse and utilize the original information of such heterogeneous data, leading to limited classification performance, and thus cannot meet the requirements for accuracy and robustness in practical applications. To this end, this paper proposes a classifier based on fuzzy similarity for mixed-attribute data. First, aiming at the mixed data structure, a fuzzy similarity adapted to the attributes is constructed, and then a fuzzy similarity measure capable of uniformly processing multiple types of attributes is built based on the product t-norm. In practical applications such as medical diagnosis, there exists a widespread problem of classifying mixed data with coexisting numerical, interval, and categorical attributes. Existing methods often fail to fully fuse and utilize the original information of such heterogeneous data, which leads to limited classification performance and thus cannot meet the requirements for accuracy and robustness in practical applications. Second, on this basis, a fuzzy similarity classifier for mixed attributes is designed to more effectively utilize the inherent structure and semantic information of the data for classification. Finally, to verify the effectiveness of the proposed classification method, it is compared with five representative classifiers, such as the novel classifier based on maximal positive region, linear support vector machine, and multi-layer perceptron. The results verify the superior performance of the new method on multiple datasets, providing an effective new approach for the mixed data classification problem.