A pruned cooperative co-evolutionary genetic neural network and its application on stock market forecast

Aiming at neural network structure designing problems, a new hybrid pruning algorithm was put forward. The algorithm consists of three steps. Firstly, it uses cooperative co-evolutionary genetic algorithm (CCGA) and back propagation algorithm (BP) to optimize the number of neural nodes and the weight values; Secondly, it calculates the significance of the hidden layer neurons; Thirdly, in order to ensure that the generalization capability of the model and simplify the network structure further, it prunes the neurons which are not significant. Using the proposed hybrid pruning algorithm to forecast stock market, simulations show that the improved algorithm has better generalization ability and higher fitting precision compared with other optimization algorithms.