Developing new physical based hybrid optimization algorithms and applications in data mining

Optimization is the process of choosing the best solution for a problem from other alternative solutions. When the problem is not linear and problem has many variables, it is difficult to solve the problem mathematically. Usually heuristic methods has been using for overcome this difficulty. Heuristic methods are very successful to find the desired solution in many problems, but they could fail obtain the desired solution in some problems. Therefore, new algorithms have been developed, or suggestions have been proposed for improve the existing algorithms. Hybridization is one of the techniques to improve the quality of the solutions of an algorithm or improve the performance of an algorithm. Two different hybrid algorithms have been proposed in this thesis. The first of these; Hybrid Electromagnetism Like - Particle Swarm Optimization - Differential Evolution (EM-PSO-DE) algorithm which is the combination of the EM, the PSO and the DE in order to improve the performance of the EM; the second; the Hybrid Parliamentary Optimization- Big Bang Big Crunch (HPO-BBBC) algorithm which is the combination of the POA and the BB-BC to improve the performance of the POA. The performances of the proposed hybrid algorithms on global optimization problems has been tested using mathematical test functions. Moreover, the performances of the proposed methods on classification have been tested with data sets from UCI and KEEL data warehouses. Performed experimental study has shown that, the hybrid EM-PSO-DE has better results than the EM, PSO and DE in global optimization problems. In the classification problems, although the hybrid EM-PSO-DE has shown slightly worse results than the EM, processing time of the hybrid EM-PSO-DE is faster than the EM. The HPO-BBBC has shown better results than the POA and the BBBC in global optimization and classification problems, however, processing time of the HPO-BBBC is slightly slower than POA.