Non-numerical nearest neighbor classifiers with value-object hierarchical embedding

摘要

Non-numerical classification plays an essential role in many real-world applications such as DNA analysis, recommendation systems and expert systems. The nearest neighbor classifier is one of the most popular and flexible models for performing classification tasks in these applications. However, due to the complexity of non-numerical data, existing nearest neighbor classifiers that use the overlap measure and its variants cannot capture the inherent ordered relationship and statistic information of non-numerical data. This phenomenon leads to the classification limitation of nearest neighbor classifiers in non-numerical data environments. To overcome this challenge, we propose a novel object distance metric, i.e., value-object hierarchical metric (VOHM), which is able to capture inherent ordered relationships within non-numerical data. Then, we construct two nearest neighbor classifiers, i.e., the value-object hierarchical embedded nearest neighbor classifier (VO-kNN) and the two-stage value-object hierarchical embedded nearest neighbor classifier (TSVO-kNN), which take advantages of both VOHM and non-numerical feature selection. Experiments show that both VO-kNN and TSVO-kNN could mine more knowledge from data and achieve better performance than state-of-the-art classifiers in non-numerical data environments. (C) 2020 Elsevier Ltd. All rights reserved.