A Novel Data Dependent Similarity Measure Algorithm Based on Attribute Selection
Nanjie Deng, Zhipeng Gao, Kun Niu · 2018
Similarity measure is an inseparable part of many data mining tasks, it possesses of great research value. Recently proposed data dependent dissimilarity measure has been proved to be more efficient than distance similarity measure in clustering, anomaly detection and multi-label classification. In this paper, we introduce a data dependent similarity measure algorithm based on attribute selection. This algorithm takes the attribute significance into consideration when building the partition model which is used to identify the similarity. Our experimental results show that this algorithm has better performance than other algorithms in anomaly detection task and is able to effectively handling high-dimensional data.