An improved genetic algorithm for optimal feature subset selection from multi-character feature set

This paper presents an improved genetic algorithm (IGA) by which the optimal feature subset can be selected effectively and efficiently from a multi-character feature set (MCFS). IGA adopts segmented chromosome management scheme to implement local management of chromosome. This scheme encodes a solution with an entire binary chromosome; but logically, it divides the chromosome into several segments according to the number of feature groups in MCFS for local management. A segmented crossover operator and a segmented mutation operator are employed to operate on these segments to avoid invalid chromosomes. The probability of crossover and mutation are adjusted dynamically according to the generation number and the fitness value. As a result, IGA obtains strong searching ability at the beginning of the evolution and achieves accelerated convergence along the evolution. IGA is tested using features extracted from cotton foreign fiber objects, and compared with the Simple Genetic Algorithm (SGA) under the same condition. The results show that IGA receives improved searching ability and convergence speed compared with SGA. The optimal feature subset selected by the IGA has much smaller size than that of the SGA. This is very important for the online classification of foreign fibers.