Improvement of RBF Neural Network by AdaBoost Algorithm Combined with PSO

Abstract

The traditional RBF neural network has the problem of slow training speed and low efficiency, this paper puts forward the algorithm of improvement of RBF neural network by AdaBoost algorithm combined with PSO, to IntroductionRBF neural network is a local optimal approximation network, with good generalization ability, classification ability and nonlinear mapping ability, simple structure, as well as fast convergence speed [1], and this network is often used in the areas, such as nonlinear system modeling, fault detection and prediction control and pattern recognition, etc [2-4]. However, there are still some problems, such as RBF neural network needs to further improve prediction accuracy, the implied layer basis function center, the appropriate number of nodes of implied layer nodes, center and width are not selected easily in the actual system, etc.Aiming at these problems, there are some algorithms to improve RBF neural network. In literature [5], it proposed using the clustering characteristics in the adaptive resonance theory network to adjust the number of implied layer nodes and the initial parameters of network, and it has simplified the structure of network, but in view of the mathematical relationship between the similarity, vigilance parameters and the width of the implied layer nodes, the network performance needs to be improved. In the literature [6], it proposed using neuron information to adjust the structure of the RBF neural network, but it did not solve the convergence problem in the process of learning. In literature [7], it proposed an improved fuzzy K-Prototype algorithm for improving RBF neural network and reducing the system sensitivity to the initial center selection of basis function, but the application scope is still limited. In literature [8], it proposed combining the gray theory with RBF neural network, but the prediction results need to improve precision. In literature [9][10][11], it proposed using PSO algorithm to optimize the parameters of RBF neural network, to reduce the error of the RBF network data.Reasonable network structure design and how to choose the appropriate learning algorithm are two main factors in the design process of RBF neural network. The determination of RBF implied layer nodes and initial parameters is the key point of the network structure design. This paper uses the improved PSO clustering algorithm to determine the initial structure of the network, then uses AdaBoost learning algorithm to learn network and train multiple weak predictors, and finally produces strong predictors, to further improve the performance of network. Finally, it uses simulation experiment of data sets from UCI database, to prove that the algorithm has improved greatly in terms of feasibility and effectiveness.

SVM Optimization Based on PSO and AdaBoost to Increasing Accuracy of CKD Diagnosis

“…The steps of the Adaboost algorithm are [23] to make predictions using the last model. Therefore, the core of the iterative AdaBoost process is iteratively AdaBoost by in circles, updating the sample to find the best weak classifier distribution at the moment, and then calculate the error rate of each weak classifier, and finally build a weak classifier into a strong classifier several times [24].…”