Improved Deep Transfer Learning Model for Scarce Sample Kechuang 50 Prediction

Aiming at the problem of limited period and scarce data samples of some financial indices, which leads to overfitting of prediction models, this paper proposes a deep transfer learning prediction model that combines random forest and LSTM. The model first applies random forest for feature selection, then inputs the filtered data to the LSTM network for training, constructs a pre-training model for source domain data, then introduces the structure and relevant parameters of the pre-training model into the transfer learning model, trains with the help of target domain data, and lastly is employed to forecast the A-share Shanghai Science and Technology Innovation 50 Index's future price trend. The experimental results demonstrate that the model provided in this study may drastically minimize the prediction error and greatly enhance the prediction accuracy of the small sample data set when compared to the comparative model. The new stock index can be predicted using the enhanced deep transfer learning model.