Research And Application On Time Series Based On Financial Data

With the wave of the Internet,more and more Internet financial companies came into being,the Internet financial risk prediction has become a priority among priorities of Internet financial companies,such as Ant Financial company predicts the inflow and outflow of Yuebao funds on the history of financial transaction time series;Melting360 campany predicts the user credit based to the user's historical repayment time series.Therefore,with the increasement of Internet financial transactions,Internet financial companies need to improve the ability to predict the risk of the system,and reduce the financial risk to a minimum.The method based on time series prediction can provide a clue for risk prediction and reduce risk.This thesis focuses on the time series prediction model based on some models,such as the traditional ARMA model,neural network model and we list the advantages and disadvantages of each model in the analysis of the application,then we propose a GT-Elman neural network prediction model based on Elman.On the other hand,we focus on the feature learning of time series,and analyze the common algorithms of feature selection and feature extraction.This thesis puts forward an internet financial risk prediction model,which is based on the combination of characteristic learning of time series data and GT-Elman neural network.The main contents are as follows:1.The application of neural network model in time series prediction is analyzed.We analyze the Elman neural network model and the model structure,analyze the characteristics of each layer of neurons,and modify the Elman neural network training algorithm.while calculating the error,we add the historical data in accordance with the current error of the distance of time assigned weights and add sequence data of stochastic process,and we put forward an improved Elman neural network model(GT-Elman)on time series prediction,so as to enhance the prediction performance of the improved Elman neural network on time series.2.The common algorithms of feature extraction and feature selection algorithms are analyzed.We transform the time domain sequence to frequency domain sequences such as fast Fourier transform and discrete wavelet transform to extract the features on the time series data;the network structure and defects of Clamping Network is analyzed in the feature selection algorithm,and we put forward an improved Clamping Network(DS-Clamping)according to the defects,and enhance the performance of the Clamping Network in the feature selection.Compared to the original data directly as the prediction model's input,the time series forecasting model can improve the prediction accuracy and improve the prediction performance of the system if we use the features which are provided by the feature learning model.3.According to the design and development of Internet financial risk prediction model system,this system is based on the SpringMVC framework,combined with Bootstrap,Echart and JQuery to build the Internet financial risk prediction system,and the corresponding prediction results are displayed visually,the result shows that this system has a good practical value.