Improved single candidate optimization algorithm for clustering problems

In order to effectively address optimization problems, the development of efficient optimization algorithms is of great importance. This thesis focuses on the Single Candidate Optimization (SCO) algorithm introduced in the literature by Shami et al. in 2022 (Shami, Grace, Burr, & Mitchell, 2022). The SCO algorithm discussed in this study is a simple and comprehensible algorithm, and its main distinction from other population-based heuristic algorithms is its attempt to find the solution to the optimization problem more quickly using a single candidate solution. However, it also faces fundamental issues such as getting stuck in local optima, similar to other heuristics. Initially, the problem of boundary violation (the candidate solution exceeding the search space) has been addressed within the SCO structure, and an accelerated opposition learning-based mechanism has been integrated into the algorithm's structure to improve search performance. Thus, in this study, a new optimization algorithm named Accelerated Opposition Learning based Single Candidate Optimization (AccOppSCO) has been proposed by incorporating the accelerated opposition learning mechanism into the SCO algorithm. To evaluate the performance of the proposed AccOppSCO algorithm, various optimization problems from the literature have been selected. The evaluation indicates that the AccOppSCO algorithm is capable of producing more accurate solutions compared to the original SCO algorithm. The performance of the proposed AccOppSCO algorithm is also evaluated on a clustering problem. In the clustering problem, the proposed AccOppSCO algorithm demonstrates superior convergence compared to the original SCO algorithm. Finally, the proposed AccOppSCO algorithm is compared with classical heuristic optimization algorithms such as Genetic Algorithm (GA), Differential Evolution (DE) and Particle Swarm Optimization (PSO). According to the results, the AccOppSCO algorithm performs better than the original SCO algorithm in terms of convergence and solution quality.