A Hybrid Particle Swarm Genetic Algorithm for Classification

The shortcomings about present genetic algorithm applying to classification are analyzed. Using the method of minimum propagating tree can cluster complex shape and non-overlap sample candidate solutions into races. The algorithm regulates optimization with "race" method and controls individuals in a micro way with race crossover. We also mixed crossover operator based on the thought of particle swarm optimization in genetic algorithm. With these operators the speed of convergence and population diversity are well balanced. Meanwhile, according to the classified question's characteristic, we designed corresponding encoding method, fitness function, and used sowing seeds way to create initial population to get better classification precision; At last, through the international data sets and classical functions, and compared with other algorithms classified effects, the results are given to illustrate the effectiveness of this algorithm.