Research On Neural Network Ensemble Based On Bagging And Generalization Ability

Neural network ensemble is the international machine learning and neural computation of a very active academic research focus. Work out a more effective neural network ensemble method to improve the neural network ensemble model generalization ability, have a very important theoretical and practical significance, it is not only conducive to machine learning and neural computation in-depth study, but also helps engineers and technicians take advantage of neural networks to solve real world problems. Neural network ensemble by training multiple neural networks and their conclusions synthesis, significantly improve the generalization ability of learning systems. The main innovation and points are as follows:(1) In Generation phase of the individual networks, BP algorithm for neural networks to learn low efficiency and the convergence speed is slow, inadequate and so easy to fall into local optimum, in order to improve the generalization ability of the individual neural networks, this paper proposes the use of genetic algorithm BP neural network of individual connections weights to optimize the method. in the genetic algorithm and neural network analysis of combined algorithm based on the proposed algorithm for improved efficiency. Algorithm code achieve in Matlab platform. And select the test data for algorithm testing. Test results show that the genetic algorithm optimization BP neural network connection weights, the convergence speed and effective solution to the BP algorithm is easily trapped into local optimization problem.(2) In this paper, the individual Bagging technology generation network, the network is generated from the individual after the results of synthesis of a key step in this integration, And for the Boston Housing Data, test to explore the integration of neural network technology in the potential applications of specific issues, abilities and prospects. Algorithm code achieve in Matlab platform. Through UCI data sets on the simulation, through the training of multiple neural networks and their synthesis of the results that can better predict the "urban per-capita crime rates" in the Boston Housing Data.