Centripetal accelerated particle swarm optimization and its applications in machine learning

• Main Author: Beheshti, Zahra
• Format: Thesis
• Language: English
• Published: 2013
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://eprints.utm.my/id/eprint/33768/6/ZahraBeheshtiPFSKSM2013.pdf; http://eprints.utm.my/id/eprint/33768/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:69883?site_name=Restricted Repository
• Institution: Universiti Teknologi Malaysia

Nowadays, meta-heuristic optimization algorithms have been extensively applied to a variety of Machine Learning (ML) applications such as classification, recognition, prediction, data mining and web mining, combinatorial optimization and so on. The majority of them imitate the behavior of natural phenomena to find the best solution. The algorithms find promising regions in an affordable time due to exploration and exploitation ability. Although the mentioned algorithms have satisfactory results in various fields, none of them is able to present a higher performance for all applications. Therefore, searching for a new meta-heuristic algorithm is an open problem. In this study, an improved scheme of Particle Swarm Optimization (PSO) based on Newtonian’s motion laws called Centripetal Accelerated Particle Swarm Optimization (CAPSO) has been proposed to accelerate learning process and to increase accuracy in solving ML problems. A binary mode of the proposed algorithm called Binary Centripetal Accelerated Particle Swarm Optimization (BCAPSO) has been developed for discrete (binary) search space. These algorithms have been employed for problems such as non-linear benchmark functions, Multi-Layer Perceptron (MLP) learning and the 0-1 Multidimensional Knapsack Problem (MKP). The results have been compared with several well-known meta-heuristic population-based algorithms in both continuous (real) and binary search spaces. From the experiments, it could be concluded that the proposed methods show significant results in function optimization for real and binary search spaces, MLP learning for classification problems and solving MKP for binary search space.