Hybridizing PSO With SA for Optimizing SVR Applied to Software Effort Estimation
Abstract: This study
investigates Particle Swarm Optimization (PSO) hybridization with Simulated
Annealing (SA) to optimize Support Vector Machine (SVR). The optimized SVR is
used for software effort estimation. The optimization of SVR consists of two
sub-problems that must be solved simultaneously; the first is input feature
selection that influences method accuracy and computing time. The next
sub-problem is finding optimal SVR parameter that each parameter gives
significant impact to method performance. To deal with a huge number of
candidate solutions of the problems, a powerful approach is required. The
proposed approach takes advantages of good solution quality from PSO and SA. We
introduce SA based acceptance rule to accept new position in PSO. The SA parameter
selection is introduced to improve the quality as stochastic algorithm is
sensitive to its parameter. The comparative works have been between PSO in
quality of solution and computing time. According to the results, the proposed
model outperforms PSO SVR in quality of solution
Keywords: particle swarm
optimization; simulated annealing; support vector regression; feature
selection; parameter optimization
Author: Dinda Novitasari, Imam
Cholissodin, Wayan Firdaus Mahmudy
Journal Code: jptkomputergg160233
