Development of new methods based on artificial bee colony programming in machine learning applications
Thesis Type: Doctorate
Institution Of The Thesis: Erciyes University, Mühendislik Fakültesi, Bilgisayar Mühendisliği, Turkey
Thesis Supervisor: Celal Öztürk
Approval Date: 2019
Thesis Language: Turkish
Abstract:
Automatic programming is a machine learning approach that tries to explain the relationship between input and output variables in a system by extracting models. Genetic Programming (GP) is the most widely used automatic programming method developed as an extension of Genetic Algorithm (GA). Artificial Bee Colony Programming (ABCP) is a newly developed high-level automatic programming method for the symbolic regression problem based on the Artificial Bee Colony (ABC) algorithm, which simulates the food source search behavior of honeybees. In this thesis, works were conducted to investigate the performance of ABCP algorithm in machine learning applications and the results were compared with GP. Furthermore, Multi-Hive Artificial Bee Colony Programming (MHABCP) and ABCP-descriptor based on ABCP algorithm were proposed for the first time as new methods. The candidate solution was generated in MHABCP used the information sharing mechanism in the standard ABCP and the adaptation of two-point high-level crossover operator used in Multi-Gene Genetic Programming to the neighborhood research process. MHABCP was used to solve the solution of high dimensional symbolic regression problems with/without noise; ABCP-descriptor was used to solve the texture classification problems. ABCP-descriptor models were extracted by sliding windows on two samples randomly selected during the training phase for the classification of each texture set. The results of this work show that ABCP and ABCP algorithm based algorithms can be used in many problems in the literature. Keywords: Automatic programming, Genetic Programming, Artificial Bee Colony Programming (ABCP), Sequence Classification, Multi Hive ABCP, ABCP-descriptor.