Stock Forecasting Based On Fractal Theory And Machine Learning

The stock market is a complex nonlinear dynamic system,with the development of quantitative investment,more and more intelligent technology is applied to the prediction of stock market,including fractal theory,swarm intelligence algorithm and machine learning are applied to the research of the stock market,and achieved certain results.However,prediction system involves attribute selection and model construction.How to use the prediction system to improve the prediction accuracy of stock market is still a hot research topic.According the previous studies about attribute selection and model construction,a prediction system combining fractal attribute selection algorithm and machine learning prediction model is proposed,and achieved a certain prediction effect.This paper first discusses the related concepts and theoretical knowledge of fractal theory,swarm intelligence algorithm,extreme learning machine and deep learning,and points out the calculation thought of fractal dimension,the calculation process of FruitFly Optimization Algorithm,the structure principle of extreme learning machine and long and short time memory network.Then a K fractal dimension method is proposed based on transformation method according the calculating thought of box counting method,which combined with FruitFly Optimization Algorithm to choose attributes,and verified this attribute selection method on the static data set and prove the effectiveness of the feature selection algorithm.Then put forward two kinds of prediction model,one model is a improved extreme learning machine combined with genetic algorithm and particle swarm algorithm,which was verified on the static data set;the other is a improved double input long and short time memory network prediction model by FruitFly Optimization Algorithm,and to verify the analysis with the Shanghai index.Finally,construct the two kinds of prediction systems combined with fractal feature selection algorithm,and use the fractal feature selection algorithm to select features of stock index,which are used to predict stock trend and yield by two prediction system,besides,combined with the back test platform according to the prediction of long and short time memory network to test the effectiveness of the prediction system.