Financial Time Series Prediction And Analysis Based On Improved RBF Neural Network

Many research achievements of the modern econometrics and finance are in need offinancial time series prediction, the prediction accuracy is often subject to variousuncertainties. Most of the financial time series are non-stationary ones, and have non-linearcharacteristics. If financial time series are predicted by the classic linear prediction model, theprediction effect is often less than ideal.RBF network can approximate any multivariate continuous function with arbitraryprecision, and the training speed is also faster, so we build financial time series predictionmodel using the RBF network. Firstly, three kinds of RBF network algorithm are introducedin this paper. Secondly, the necessity of improving RBF neural network width is illustrated bycomputer experiments, then a search algorithm of RBF neural network width is proposed, anda detailed description of the process of the entire algorithm is given. Finally, the effectivenessof this algorithm is verified by the simulation results.Neural network ensemble technology is a research hot topic in recent years, combiningthe predicted outputs of multiple neural networks into an aggregate predicted output oftengives improved accuracy over any predicted output of individual neural networks. Theresearch on neural network ensemble technology is mainly concentrated on two aspects, oneis how to generate the individual neural network, the other is how to aggregate neural network.Firstly, the reason for the RBF network combination method need to be improved is given inthis paper. Secondly, the improvement of the combination algorithm of the RBF Neuralnetwork ensemble is presented. Finally, the purchasing managers’ index prediction resultsshow that this improvement is effective.This article main research results are as follows:(1) A method searching for RBF network width is presented, the experiments show thatthe fitting effect of the improved width RBF network is better than that of fixed-width RBFnetworks.(2) Final prediction accuracy is greatly influenced by the abnormal predictive error ofindividual RBF network output in the process of RBF neural network ensemble prediction.The predictive value of individual RBF network model with the help of SAS statisticalanalysis software is analyzed, and the exceptional prediction values are removed out, then therest of the RBF network predicted output are combined. The results show that this method canimprove the prediction accuracy of RBF network ensemble prediction model...