Preface

This book is about how to use new algorithm models to solve complex problems. The book presents one branch of a field in computer science that we today call computational intelligence. Big Data play already a great role in society and evolutionary algorithm may be one approach to do data mining of high-dimensional data. High-dimensional data are produced in scientific laboratories all over the world and are often difficult to interpret. Clustering is one possible technique that may help us to interpret these data.

Clustering is a well-known technique that is used in many areas of science. This book is about how to use such algorithms to solve clustering of huge sets of data. This subject may be introduced in the last year at the bachelor level in computer science or mathematics or at the graduate level. The intention of the book is to show how to use such computation models on classical clustering problems. Visualization is an important part of the clustering process. We therefore also want to visualize the result of the cluster analysis.

The most known clustering algorithm is the K-means algorithm that is dependent on the parameter value K and the initial position of K cluster centroids. An incorrect value will result in an inaccurate clustering structure. The configuration of cluster centroids determines if the algorithm converges to a local minimum or not. These limitations may be solved by using evolutionary algorithms.

Genetic algorithms and differential evolution algorithms are two paradigms in the book that are used to optimize the value of K and the initial configuration of cluster centroids. The correctness and quality of the solution are compared using both artificial and real-life data sets. Experiments have shown that the algorithms are able to classify the correct number of well-defined clusters, but fail to do so for overlapping data clusters. This is mainly because the Davies-Bouldin Index as a fitness measure has certain kind of limitations. The experiments carried out in the book also show that both Genetic and Differential evolution algorithms provide suboptimal positions of initial configuration of cluster centroids, reflected in higher values of the Davies-Bouldin Index.