Automatic Teaching–Learning-Based Optimization: A Novel Clustering Method for Gene Functional Enrichments - Computational Intelligence Techniques for Comparative Genomics - page 17

Keywords Automatic clustering Teaching learning-based optimization Gene

-

functional enrichments Cluster validity indices

1 Introduction

Evolutionary algorithms (EA) are generic meta-heuristic optimization algorithms
that use techniques inspired by nature s evolutionary processes. EA maintains a

whole set of solutions that are optimized at the same time instead of a one single
solution. The inherent randomness of the emulated biological processes enables
them to provide good approximate solutions nevertheless. The recently emerged
nature-inspired multi-objective meta-heuristic optimization algorithms teaching

'

-
learning-based optimization (TLBO) [ 1 , 2 ] and its variations Elitist TLBO [ 3 , 4 ]
belong to this category. Both these algorithms aim to

-

nd global solutions for real-
world problem with less computational effort and high reliability. The principle idea
behind TLBO is the simulation of teaching

learning process of a traditional
classroom in to algorithmic representation with two phases called teaching and
learning. Elitist TLBO was pioneered with a major modi

-

cation to eliminate the

duplicate solutions in learning phase.
Clustering is the subject of active research in several elds such as statistics,

pattern recognition, machine learning, data mining, and bioinformatics. The pur-
pose of clustering is to determine the intrinsic grouping in a set of unlabeled data,
where the objects in each group are indistinguishable under some criterion of
similarity. Clustering is used to partition a dataset into groups, so that the data
elements within a cluster are more similar to each other than data elements in
different clusters. Automatic clustering addresses the challenge of determination the
appropriate number of clusters or partitions mechanically.
Most of the existing clustering techniques, based on EA, accept the number of
classes (k) as an input instead of determining the same on the iteration. Never-
theless, in many practical situations, the appropriate number of groups in a previ-
ously unhandled dataset may be unknown or impossible to determine even
approximately. To avoid the algorithm struck in such blockage, automatic assign-
ment of (k) value by the algorithm in each run was made tangible in this work.
These automatic clusters are again endorsed with cluster validity indices (CVIs),
which combine compactness and separability for assessing the quality of clusters.
Cluster validity criteria are of three types external, internal, and relative. External
indexes require a priori data for the purposes of evaluating the results of a clustering
algorithm, whereas internal indexes do not. Internal indexes evaluate the results of a
clustering algorithm using information that involves the vectors of the datasets
themselves. The relative index evaluates the results by comparing the current
cluster structures with other clustering schemes. The CVIs that are used in this work
are rand index (RI) [ 5 ], advanced rand index (ARI) [ 5 ], Hubert index (HI) [ 6 ],