Abstract:

The traditional BP (Back Propagation) neural network is easy to fall into local minimum and has lower accuracy. According to this problem, a method that combines the Adaboost algorithm and BP neural network is proposed to improve the prediction accuracy and generalization ability of the neural network. Firstly, the method preprocesses the historical data and initializes the distribution weights of test data. Secondly, it selects different hidden layer nodes, node transfer functions, training functions, and network learning functions to construct weak predictors of BP neural network and trains the sample data repeatedly. Finally, it made more weak predictors of BP neural network to form a new strong predictor by Adaboost algorithm. The database of UCI (University of California Irvine) is used in experiments. The results show that this method can reduce nearly 50% for the mean error absolute value compared to the traditional BP network, and improve the prediction accuracy of network. So this method provides references for the neural network prediction.

Key words: neural network;strong predictor;iterative algorithm;adaboost algorithm