Optimizing neural network classification by using the Cuckoo algorithm

Classification has become a very important field in the current era of big data. As one of the main stream algorithms for classification, the well-known Error Back Propagation algorithm, with the characteristic of nonlinear mapping, good self-study ability and fault tolerance ability, has been pervasively applied in finance, agriculture, industry and other fields. However, the Error Back Propagation algorithm would face the problems of low accuracy, poor stability and slow convergence if the weights and thresholds are set improperly. In this paper, the Cuckoo algorithm is employed to train the weights and thresholds of the Error Back Propagation algorithm. From the aspects of accuracy, stability and time cost, experiments and performance comparisons towards the basic Error Back Propagation algorithm model (BP), the improved neural network model based on Cuckoo algorithm (BPCS) and the improved neural network model based on Genetic algorithm (GABP) are organized by using two classification datasets, respectively. The results show that the neural network optimizing by Cuckoo algorithm has faster convergence speed, higher accuracy and better stability than others. In addition, the ranges for selecting parameters are suggested based on an appropriate model.