Theoretical Utility of Data Value Metric and Genetic Algorithms for Variable Clustering in an Unsupervised Learning Environment

Abstract

Cluster analysis is regarded as one of the most important unsupervised
learning tasks, with its natural application in dividing data into meaningful
groups, also known as clusters, based on the information in the data by
describing the objects in terms of their relationships and capturing the data's
natural structure. Many traditional performance evaluation metrics for
clustering algorithms abound in the literature, treating various attributes or
variables equally when measuring similarity; however, different attributes or
variables may contribute differently due to the amount of information they
contain, which can vary greatly. Data Value Metric (DVM) is an information
theoretic measure based on the concept of mutual information that has been
shown to be a good metric for validating data quality and utility in a big data
ecosystem and in traditional data. Because it uses a forward selection
search strategy, Data Value Metric (DVM) suffers from local minima and
loss of diversity in the population; however, hybridizing it with Genetic
Algorithm will overcome the problem of local minima because there will be a
blend of evolutionary search to ensure a balance between exploration and
exploitation of the search space. This paper proposed a hybrid model of the
Genetic Algorithm and the Data Value Metric (DVM) as an information-theoretic metric for quantifying the quality and utility of variable clustering
selection that can be applied to traditional data.