الفهرس 
1 IntroductionOnly 14 pages are availabe for public view
 |  | from | 99 |  |  |
| | | from | 99 | | |
Abstract
Swarm intelligence (SI), bio-inspired computation in general, has attracted great interest
in the last two decades, and many SI-based optimization algorithms have gained huge
popularity. There are many reasons for such popularity and attention, and two main reasons
are probably that these SI-based algorithms are flexible and versatile, and that they are very
efficient in solving nonlinear design problems with real-world applications.Bio-inspired
computation has permeated into almost all areas of sciences, engineering, and industries,
from data mining to optimization, from computational intelligence to business planning, and
from bioinformatics to industrial applications. In fact, it is perhaps one of the most active
and popular research subjects with wide multidisciplinary connections. In this thesis, we try
to introduce a novel binary Whale Optimization algorithm, which we called bWOA-S and
bWOA-V using transfer function. WOA is one of the recently meta-heuristics algorithms
that mimic the behavior of humpback whales to accomplish global optimization. Moreover,
WOA is operating mainly in continuous search space. Binary whale optimization algorithm
introduced here is performed to map a continuous search space of WOA to binary search
space via transfer function was employed.
In this thesis, we try to introduce a novel binary Whale Optimization algorithm, which
we called bWOA-S and bWOA-V using transfer function. WOA is one of the recently
meta-heuristics algorithms that mimic the behavior of humpback whales to accomplish
global optimization. Moreover, WOA is operating mainly in continuous search space. To test
our algorithm mathematically, we run it on 22 benchmark function CEC 2005 and take the
average of 30 run. Then we test it on 3 of engineering problem. Before applying our binary
algorithm to chemical dataset,we firstly applied it on Feature selection Problem.
Feature selection (FS) was touted as global combinatorial optimization, aim to simplify
and enhance the quality of high-dimensional datasets by selecting prominent features, removing
irrelevant, and redundant data to provide good classification results. The objectives of
FS are dimensionality reduction and improving the classification accuracy generally utilized
with great importance in different
vi
fields such as pattern classification, data analysis, and data mining applications. In order
to overcome the curse of dimensionality problem, our two novel binary variants of the
whale optimization algorithm (WOA) are proposed and used to decrease the complexity
and increase the performance of a system by selecting significant features for classification
purposes, then we select 24 benchmark dataset from UCI repository.
To test our algorithm we select 24 of different size dataset from UCI repository and we
compare our algorithm with the state of the art algorithms such as Particle Swarm Optimization(
PSO) and Genetic Algorithm(GA) and three instant of binary ant lion Optimizer bALO1,
bALO2 and bALO3. we also test it with the native algorithm and binary Dragonfly Algorithm
(bDA). we run our experiments 20 run, in each run 70% of dataset selected randomly as
Training Data and the other as test. We used a set of Criteria to test and compare these
algorithms.Eventually, Wilcoxons rank sum nonparametric statistical test was carried out
at 5% significance level to judge whether the results of the proposed algorithms differ from
those of the other compared algorithms in a statistically significant way. The quantitative and
qualitative results revealed that the two proposed variants in the FS domain are capable of
minimizing the number of selected features as well as maximizing the classification accuracy
within an appropriate time.
Chemoinformatics (chemical informatics) is the field that seeks to use informational
techniques like computer science, mathematics and information techniques to predict or
analyze molecule’s (chemical compounds) properties. One of the major principles in this
research field is the similarity principle, which states that two structurally similar molecules
should have similar activities and properties. Quantitative Structure Activity Relationship
(QSAR) is a field which is based on finding the correlation between molecule’s descriptors
and molecule’s properties.