K-Means Clustering with Missing Data

• Activity Number: 660
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistical Computing
• Author(s): Juwon Song*
• Companies: Korea University
• Keywords: k-means clustering ; missing data ; imputation

Abstract

K-means clustering is one of the most popular analysis techniques to divide observations into several groups. K-means clustering is simple and easy to implement for completely observed data. When data include missing values, it is often to impute missing values in the preprocessing stage of k-means clustering, but it would ignore membership information when we impute missing values. To overcome this disadvantage, it has been suggested to impute missing values using the observed mean or nearest neighbors when we calculate distances between observed values and group means. On the other hand, the K-means clustering can be considered as a special case of clustering based on finite mixture models. We consider an extension of the finite mixture model by using the EM algorithm to handle missing data in K-means clustering. A simulation is conducted to evaluate the performance of the suggested method and compare it with other commonly used approaches to handle missing data in K-means clustering.