A new hybrid algorithm for feature selection and its application to customer recognition

Abstract

This paper proposes a novel hybrid algorithm for feature selection. This algorithm combines a global optimization algorithm called the simulated annealing algorithm based nested partitions (NP/SA). The resulting hybrid algorithm NP/SA retains the global perspective of the nested partitions algorithm and the local search capabilities of the simulated annealing method. We also present a detailed application of the new algorithm to a customer feature selection problem in customer recognition of a life insurance company and it is found to have great computation efficiency and convergence speed. © Springer-Verlag Berlin Heidelberg 2007.