| The prediction of time series becomes a hot issue in recent years.There are many time series in real life,such as temperature,hydrology,stock index,they are relevant to everyday life.One can prepare for future events when the time series be predicted accurately.This thesis focuses on the prediction of CSI 300 index.CSI 300 Index is an important barometer of the financial market.It is one of the most representative indexes of China’s stock market and can reflect the overall situation of China’s stock market.If we can predict the CSI 300 index accurately,we can deal with the risks in the future.This thesis collects 8 basic data of CSI 300 index and 22 foreign market index from 2013 to 2023.These data are trained by LSTM model,so that the model can predict the index of the next day through the data of the previous 5 days.Because each market closes at different times,the time series is no longer continuous.In this thesis,an optimized ARIMA algorithm is proposed,which is used to fit the missing value and replace the missing value with the fitting value to keep the time series continuous.Then,four data processing methods,correlation analysis,PCA,ARIMA and classification standardization,were selected to process the data.In this thesis,different methods are used to combine different models.After horizontal comparison between different models,comparative analysis is made to analyze the advantages and disadvantages of different models,and the optimal model is given.Finally,the model is used to predict different predictive variables.Then this thesis analyzes the effects of different predictive variables on the final results.The innovation of this thesis is as follows:1.Propose the improved ARIMA algorithm,2.Different combination models are constructed with different data processing methods,3.Compare and analyze different prediction variables and reveal the importance of predictive variables.It can be seen from the subsequent prediction results that when the computational power is limited and the depth of the model is limited,the influences on the model results are ranked from greatest to smallest,which are respectively the selection of prediction variables,data dimension and data processing.Therefore,in order to further improve the prediction accuracy of the model,we should first consider the selection of appropriate prediction variables,introduce more dimensions of data,and finally focus on data processing.That means we need more financial knowledge than just optimizing machine learning models.Machine learning models are great predictive tools,but they are not the solution. |
