An incremental mixed data clustering method using a new distance measure

Abstract

Clustering is one of the most applied unsupervised machine learning tasks. Although there exist several clustering algorithms for numeric data, more sophisticated clustering algorithms to address mixed data (numeric and categorical data) more efficiently are still required. Other important issues to be considered in clustering are incremental learning and generating a sufficient number of clusters without specifying the number of clusters a priori. In this paper, we introduce a mixed data clustering method which is incremental and generates a sufficient number of clusters automatically. The proposed method is based on the Adjusted Self-Organizing Incremental Neural Network (ASOINN) algorithm exploiting a new distance measure and new update rules for handling mixed data. The proposed clustering method is compared with the ASOINN and three other clustering algorithms comprehensively. The results of comparative experiments on various data sets using several clustering evaluation measures show the effectiveness of the proposed mixed data clustering method.