Research On Financial Data Analysis Method Based On Deep Learning

In the era of information technology development in the financial field,the amount of financial data has increased dramatically.Many investors and researchers try to obtain some effective decisions and help through in-depth analysis of these data to minimize the losses caused by decision-making mistakes.Therefore,the research on financial data analysis method is a very useful and hot topic at present.Among them,the research on the imputation of missing financial data and the prediction of financial time series are two more important modules.Currently,deep learning,with its powerful ability to capture data hiding patterns and underlying dynamics,has shown excellent results in data analysis tasks in various fields.However,in the face of complex financial data,the effect achieved by simple application of deep learning model can not meet the actual needs.Therefore,this paper focuses on the imputation of missing financial data and the prediction of financial time series,and aims to improve the existing deep learning model to obtain better performance.The specific work of this paper is as follows:(1)Aiming at the problem of missing financial data,a novel imputation model named as Multiple Generative Adversarial Imputation Networks(MGAIN)is proposed.Specifically,we first randomly select multiple attribute subsets instead of the whole attributes such that more complete samples can be generated.Then,the missing data in each attribute are imputed by using Generative Adversarial Imputation Networks(GAIN),which fully considers the relationships among missing values by combining neural network and adversarial learning.The proposed subset selection and multiple imputation strategy not only simplifies the network structure of GAIN,but also reduces the dependence of model performance on data.Finally,a weighted average method is presented to synthesize multiple results of each missing attribute value to further improve the accuracy.The experimental results based on bank credit risk assessment data and stock price collapse risk data demonstrate that the proposed method has achieved a good imputation effect and is superior to some commonly used missing data imputation methods.(2)Aiming at the problem of financial time series trend prediction,a time series prediction model based on Personalized Simple Recurrent Unit(PSRU)is proposed.Specifically,we firstly introduce Variational Mode Decomposition(VMD)in terms of parameter determination for denoising.Then,to better capture the characteristics of diverse time series,we cluster the time series into multiple classes,and each class fits an optimal prediction model.Especially,to improve the clustering,we introduce the Dynamic Multi-perspective Personalized Similarity Measurement(DMPSM),which can help solve the problems of singularities,time shifts and warpings and personalized characteristics in time series.Finally,Simple Recurrent Unit(SRU)is adopted as the prediction model,which can achieve high prediction accuracy and speed with excellent mapping and parallel processing capabilities.The experimental results based on the stock data from the Shanghai Stock Exchange demonstrate that the prediction effect of our proposed model is significantly better than some popular time series prediction methods.