| With the development of statistics and computer science,quantified investment has been popular.From statistical models to deep learning models,new financial tech-nology models and quantitative strategies are constantly enriched.According to the data characteristics and data labels in finance,people keep training the model to fit the relationship and try to predict it,and the model will be appied to the market once the prediction is accuracy enough.Till now,the training method of the model is mainly based on the supervised training of label.In the traditional supervised deep learning,LSTM neural network has been proved to have a certain ability to grasp and predict the characteristics of financial data.However,because of the high noise and low informa-tion ratio in the data,LSTM model also relies too much on the absolute value or label given by the data,and has great impacts on the prediction.In order to solve the problem on the data and the traing,this paper takes a new per-spective,semi supervised learning,and tries to compare supervised training with semi supervised training,this method can not only be applied into financial data analysis,it can also be applied into image analysis etc.In specific,we use GAN(general advertis-ing network)network,a Semi-supervised learning mode which uses networks training networks,to handle the classification of the rising or falling price.Through GAN net-work,some virtual data is produced to assist in the completion of classification tasks,reducing the model’s dependence on data label.The discriminant network of GAN net-work is set as LSTM neural network.The data analysis shows that the LSTM neural network based on the Semi-supervised GAN network training has better performance compared to the traditional supervised LSTM neural network in quantitative trading. |
