An Application of Genetic Algorithm for Clustering Observations with Incomplete Data

Abstract

Cluster analysis is a method to classify observations into several clusters. A common strategy for clustering the observations uses distance as a similarity index. However distance approach cannot be applied when data is not complete. Genetic Algorithm is applied by involving variance (GACV) in order to solve this problem. This study employed GACV on Iris data that was introduced by Sir Ronald Fisher. Clustering the incomplete data was implemented on data which was produced by deleting some values of Iris data. The algorithm was developed under R 3.0.2 software and got satisfying result for clustering complete data with 95.99% sensitivity and 98% consistency. GACV could be applied to cluster observations with missing value without any pre-processing step to incomplete observations. Performance on clustering observations with incomplete data is also satisfying but tends to decrease as the proportion of incomplete values increases. The proportion of incomplete values should be less than or equal to 40% to get sensitivity and consistency not less than 90%.