Analysis And Research Of The Stock Split Algorithm Based On Neural Network

With the development of artificial intelligence technology and theory,the financial sector has begun to undergo tremendous changes.The huge volume of transactions and the accuracy and completeness of historical datas have made the financial sector more procedural,systematic and wise.VWAP(Volume Weighted Average Price),as a popular single splitting algorithm in the financial sector,which can split large orders,reduce the impact of big orders on the stock market,and improve the concealment of large orders.However,with the diversification of trader's goals and the continuous transformation of the trading market,the traditional VWAP algorithm is no longer stable,and it cannot meet the needs of the public.It's time to find new solutions.Considering that in recent years,the deep neural network algorithms have made significant achievements in many fields,especially the long-short term memory(LSTM)neural network can be said to be the preferred neural network for dealing with time series.Therefore,this paper introduces LSTM neural network algorithm to reconstruct the stock model of split orders.1)This paper analyzes the implementation process of the traditional VWAP algorithm and its disadvantages.It designs a stock split strategy for short-term high-frequency trading.According to the different forecasting requirements in the split strategy,the corresponding regression prediction model and classification prediction model based on the LSTM neural network were constructed respectively.In this paper,different performance evaluation indicators were used to analyze and verify the prediction effect of the two models.It is concluded that the performance of the neural network regression model for stock price forecasting far exceeds the classification forecasting model for forecasting the upward and downward trend of prices.2)By reading a large number of related literatures,a set of feature sets including the factors of price fluctuation,volatility,and inventory pressure are defined and calculated.Through the comparative analysis of the performance of the model under the combination of different characteristic factors,the influence of different characteristic factors on the stock prediction can be observed.3)Considering that the training data set is too large,a stochastic gradient descent method with high training efficiency is used to train the LSTM network model.By setting different training parameters such as the number of hidden neurons,the learning rate,and inputting Number,training times,etc.to observe the training effect of the model.Through continuous optimization of parameters,a set of better-performing training parameters was finally found,so that the model can better avoid under-fitting and over-fitting,showing a good prediction effect.The excellent predictive effect confirms the applicability of LSTM neural network in algorithmic trading and lays a solid foundation for the development of neural network algorithms in the financial field.