Research On The Prediction Of Market Index Based On Ensemble Learning And Multi-Angle Analysis

With further extension and development of information technology,data from various industries has increased exponentially.How to effectively utilize the data to seek the regular patterns,control and prevent the loss in human life and society has become a current research hotspot,including the research of forecasting of financial data.The financial market is a complex,non-stationary and non-linear system,thus it has always been the mission and responsibility of the government and financial regulatory departments to stabilize the financial market and promote the healthy development of social economy.As a focus in the quantified investment area,the theory&application of stock forecasting has been enriched and improved.A sound stock forecasting system can provide feasible suggestions for investors,which promotes the virtuous cycle of capital,and it also helps the government to stabilize the development of market economy.In this paper,the information of stocks is explored from multiple layers based on the daily K-line stock data and prediction outcomes of stock price in future days have achieved good performance.In terms of data preparation,this paper considers historical closing price time series data and K-line trading data;in terms of model construction,it integrates one statistical model,one machine learning model and neural networks;in terms of model evaluation,it introduces a number of indicators,and carries out trend and inflection point analysis in combination with K-line theory.First of all,it uses the signal processing technique,singular spectrum analysis(SSA)to reduce the noise of time series,so as to retain its primary components.Next,it combines the statistical model-autoregressive moving average(ARIMA)and a variety of neural networks together to forecast the future closing price of stocks.At the same time,it chooses the machine learning model,gradient boosting regression tree(GBDT)to conduct the nonlinear regression prediction,so as to make the best of the information from the stock data.Next,based on the idea of stacking methods in ensemble learning,a neural network distinguished from previous ones called extreme learning machine is selected as final predictor which collects the forecasting results from individual models and outputs the ensemble prediction results in the end.The forecasting system effectively utilizes the advantages of each model and spontaneously makes up for their shortcomings.It then introduces verification of precision and stability of the forecasting system.The selected error indexes include mean absolute error(MAE),root mean square error(RMSE),mean absolute percentage error(MAPE)and Theil inequality coefficient(TIC).Next,to measure the validity of the model,this paper selects the effectiveness index to quantify the comparative analysis of the model effectiveness.It compares the differences of several error indexes between the extreme learning machine(ELM)ensemble strategy and the simple combination average method.Trend chart,error histogram,scatter chart of prediction value and actual value,error fluctuation curve are introduced in this paper as well for above verification analysist to intuitively present the advantages of the ensemble model in this paper.Finally,in comparative analysis,it also combines the K-line chart and K-line theory with the prediction curve.In this paper,the research of stock forecasting system has obtained the following conclusions:First,the ensemble model achieves the highest forecasting precision in all comparisons of error indexes in all stock markets,among which the accuracy improvement efficiency of Shanghai Composite Index is the highest.In the error fluctuation chart,the range of error of the ELM ensemble model is the smallest.When observing the scatter diagram between the predicted value and the real value,the figure is close to a straight line passing through the origin.According to the probability distribution diagram of the error,the error distribution of the predicted value is close to the normal distribution.Therefore,the ELM ensemble model has good fitting,high precision and stability.In addition,according to the effectiveness of the forecasting system,the ensemble model has passed the first-order and second-order effectiveness test;in comparison with the combination strategies,it shows that the integration strategy of the ELM ensemble model has more advantages than models of general combination average.Combined with K-line chart and K-line analysis,the ensemble model is easier to approach the top of K-line chart in the soaring period of stock,and closer to the bottom of K-line chart in the period of stock price falling,and also has high sensitivity at the turning point.In conclusion,this paper believes that ELM ensemble forecasting system breaks through the limitations of empirical interpretation of traditional technical indexes,and makes up for the inherent structural defects of individual models.It can maximize the prediction performance of the stock forecasting system by fitting and analyzing multi-layer stock data with the following novel ideas:? In the time series model,signal processing technique is introduced to filter the high frequency information and improve the prediction accuracy;? all of weak supervised models can simultaneously capture the linear and nonlinear components in the stock information;?the ensemble learning technique could absorb each models'advantages and improve its fault tolerance and stability;? a more comprehensive evaluation of the market index forecasting system is introduced by combining K-line analysis.Nevertheless,there are some deficiencies in the prediction system,such as the inability to ensure the independence of training samples,and it cannot provide good forecasting results under the influence of emergencies.On the whole,the large market forecasting system proposed in this paper can achieve both high prediction accuracy and strong robustness,providing guidelines for the government to create a good investment and financial environment.