Research On Stock Forecast Based On The Improved Method Of Machine Learning

People's living standards in our country are constantly improving with the rapid development of economy.The demands for investment are growing and the methods are becoming variety as people become prosperity,stock investment is a traditional and widely used way to realize.Once an investment strategy is developed based on an efficient stock price forecasting method,the rise will decrease and the income will increase.In the paper,two machine learning algorithms are used to forecast the stock price,and the results are used to formulate individual and portfolio investment strategies respectively.Support vector regression machines and RBF neural network models are widely used in solving regression problems,especially when dealing with economic data.Based on the previous research results,the paper applies these two machine learning algorithms to solve the actual problem of stock price forecasting.During the experiment,the sequence of closing stock price is used as the basic data,and the normalized closing price vector is reconstructed by the coordinate delay method.The parameter time delay of vector reconstruction and matrix dimension are determined by complex autocorrelation method and G-P algorithm.The experiment using genetic algorithm GA to optimize the regularization parameters and kernel function width of support vector regression machines SVR and LSSVR.Using K-means clustering algorithm to optimize the data center point and expansion constant of the Gaussian radial basis activation function of the hidden layer of RBF neural network,calculating the initial weight of the network by using the cluster number as the number of hidden layer neurons.The paper forecasts the closing stock price based on these two algorithms.Using the coordinate delay method to reconstruct the time series of closing stock price simplifies the work of selecting the data dimension and enables the model to forecast the closing stock price based on historical data.After optimizing the parameters of support vector regression machine and RBF neural network,the experiment forecasts the closing price of eight stocks at different time based on these two algorithms,the results are shown in the paper.Through the analysis of the prediction results of the model and evaluated the evaluation index,the conclusions show that the support vector regression machine has a better fitting effect on the training set than the RBF neural network model,while the RBF neural network model has a slightly better prediction effect on the prediction set than the support vector regression machine.Both models forecast the trend of the closing price effectively.Simulated investment experimentations were made using different strategies for partial stocks with rising closing prices,the results of the yield further validate the feasibility of the stock price forecasting method.