Application of MEA-LSTM Neural Network in Stock Balance Prediction

Abstract

AbstractIn the financial industry, the effective control of stock data has been concerned by various departments in this field. The stock is affected by a large number of controllable and uncontrollable factors, it is difficult to predict the future development, so it needs the help of computer and data. However, in the past research, the gradient vanishing problem will appear when using the recurrent neural network, the long short term memory network (LSTM) neural network is prone to over fitting, when the weights and thresholds are randomly generated by the network itself or artificially set according to the summary of many experiment. Therefore, in view of the characteristics of stock data, this paper proposes a model, which uses the mind evolutionary algorithm to optimize the long-term and short-term memory cycle memory neural network, so it can do the maximum degree of data fitting. This model not only retains the characteristics of LSTM neural network, but also alleviates the over fitting problem. It can avoid the situation that the gradient of recurrent neural network (RNN) network will disappear because of too long time series, and also relatively alleviate the problem of over fitting of traditional LSTM neural network. Finally, the comparison between LSTM and mind evolutionary algorithm-long short term memory network (MEA-LSTM) models shows that the prediction effect of our proposed model is better than the traditional LSTM.KeywordsData predictionNeural networkMEA-LSTM