Attribute selection using binary flower pollination algorithm with greedy crossover and ‘one to all’ initialisation

Abstract

Wrapper-based methodology for attribute selection is achieved by employing ‘support vector machine (SVM)’ and ‘binary flower pollination algorithm (BFPA)’. A greedy crossover is proposed to reset the suboptimal solution obtained on pre-mature convergence. Also, ‘one to all’ initialisation is developed to devise the initial pollen population for diversified exploration. The proposed methodologies are validated with datasets from uc irvine (UCI) repository and results show superior performance on comparison with the literature on both ‘BFPA’ and other metaheuristic algorithms for attribute selection.