A novel approach to solution of data science and engineering optimization problems: Chaotic artificial algae algorithm

Optimization is the process of finding the optimal solution in the solution space of a problem and determining the best one under the given constraints. In today's world, it is widely used in many areas where maximum efficiency is aimed with minimum cost. In recent years, the increasing complexity and difficulty of real-world problems has led to a greater need for more reliable optimization techniques, especially metaheuristic optimization algorithms. Artificial Algae Algorithm (AAA) is a metaheuristic optimization algorithm inspired by the characteristics and life behavior of microalgae. It has become one of the popular metaheuristic algorithms by successfully solving many real-world problems in various fields. However, similar to other metaheuristic optimization algorithms, AAA tends to early converge and get to stuck in local minima. In order to overcome these problems, the structure of the algorithm needs to be strengthened. There are two important search strategies that determine the characteristics of metaheuristic optimization algorithms. One is exploration/diversification and the other is exploitation/intensification. The discovery process is the ability to explore the search space globally. This ability is the ability to avoid the local optimum and get rid of the local optimum when stuck. The exploitation process is the ability to discover possible solutions near the current solution in order to locally improve the relevance of the solution. The perfect performance of a metaheuristic optimization algorithm depends on the balance between these two strategies. Various strategies such as levy flight, quantum behavior, local search, multiple and intelligent search, chaos theory have been developed in the literature to strengthen the exploration and exploitation processes and to create the balance between them. One of these strategies is chaotic maps, which are inspired by chaos theory. Chaotic maps are a very important performance enhancement strategy that strengthens the balance between exploration and exploitation, as the performance of many metaheuristic optimization algorithms in the literature has been improved by using these maps. In this thesis, a new approach named Chaotic Artificial Algae Algorithm has been developed by equipping AAA with chaotic maps and it has brought a solution to four different problem spaces. With this developed approach, first of all, thirty benchmark test functions of different difficulty were solved. The performance was then validated by testing on the pressure tank design, welded beam design, tension compression spring design, and eight space design problems from the European Space Agency. Thirdly, the chaotic AAA approach is applied to the distributed clustering problem in unsupervised learning, which is one of the three main areas of machine learning, and its performance is analyzed. Fourth, a binary version of the proposed chaotic AAA approach has been developed to be used in feature selection, which is an inevitable critical preprocessing process for machine learning algorithms. The chaotic-based new approach developed in this thesis has been compared with various popular algorithms on problem sets with different difficulty levels in the literature in all problem spaces, its reliability has been ensured by performing Wilcoxon signed-rank test and Friedman statistical tests and it has been verified that it is more performant than the compared algorithms.