Nonparametric Clustering of Mixed Data Using Modified Chi-square Tests

TL;DR

This paper introduces a non-parametric clustering method for mixed continuous and discrete data using modified Chi-square tests, avoiding the need for a global distance function, and demonstrates superior performance over existing methods.

Contribution

The paper presents a novel non-parametric clustering approach for mixed data that leverages local neighborhood analysis and modified Chi-square tests, without requiring a global distance metric.

Findings

Outperforms AutoClass in simulation studies

Effective for various data settings

No need for a global distance function

Abstract

We propose a non-parametric method to cluster mixed data containing both continuous and discrete random variables. The product space of continuous and categorical sample spaces is approximated locally by analyzing neighborhoods with cluster patterns. Detection of cluster patterns on the product space is determined by using a modified Chi-square test. The proposed method does not impose a global distance function which could be difficult to specify in practice. Results from simulation studies have shown that our proposed methods out-performed the benchmark method, AutoClass, for various settings.