Research And Realization Of Ensemble Learning Method For Stock Trend Prediction

The research of stock trend prediction is an important research direction of financial big data.With the development of artificial intelligence technology,Stock trend prediction has expanded from the traditional basic analysis methods to the use of machine learning methods for technical analysis.Among them,algorithms of Neural Network and Ensemble Learning provide new modeling solutions for stock trend prediction.But base learners of traditional Ensemble Learning only use decision tree or LSTM neural network,which has certain limitations.The main contents are as follows:1)Design an LSTM model that superimposes the attention mechanism and adversarial learning: AALSTM model.Because stock data is non-linear,non-stationary,and contains a lot of noise,it is difficult for ordinary LSTM models to capture potential features.In order to solve this problem,this paper designs an LSTM neural network model with superimposed attention mechanism and adversarial learning to model and predict stock data.By comparing the prediction effect of the AALSTM model and the traditional LSTM model,the experimental results show that the prediction ability of the AALSTM model is higher than that of the traditional LSTM model.2)Design an ensemble learning model based on the AALSTM model : BagAALSTM model.Aiming at the problem that the generalization ability of the AALSTM single model is limited,and the prediction ability cannot be further improved.This paper use the Bagging parallel Ensemble Learning to integrate the AALSTM model as its base classifier.By comparing the prediction effects of the BagAALSTM model and the single AALSTM model,the experimental results show that the prediction ability of the BagAALSTM model is better than that of the single AALSTM model.3)Using genetic algorithms for feature selection.In order to avoid too many features to cause the model to overfit training data.In this paper,genetic algorithm is used to screen out the optimal feature subset suitable for BagAALSTM model.4)Building a distributed Ensemble Learning model.To solve the problem of real-time feedback of model prediction results in stock scenarios,Provide computing services for the BagAALSTM Ensemble Learning model.When there are many Base classifiers,it can provide more stable,safer and faster computing services than a centralized architecture.