Determine of appropriate neural networks structure using Ant Colony System

This paper proposed a hybrid training algorithm by combining the Ant Colony System and BP algorithm. The Ant Colony System is used optimize the initial of the BP neural networks structure, connection between neurons and connection weights. The yield structure has trained using BP algorithms. This method can cope with trapping local minimum problem of the BP algorithm. The proposed method and the standard BP neural networks are applied to pattern classification problems from PROBEN1 benchmark data set, and we chosen Breast Cancer data set for our experimentation. The results show that the precision and efficiency of NN structure from the proposed method are better than the standard BP neural networks.