Deep Learning in Stock Price Prediction

Predicting stock prices is a crucial field of financial research. Traditional time series forecasting method have limitations in capturing non-linear features. It's difficult to identify the complex linkages embedded in the data. In this study, Xiaomi's stock data (2020-09-01 to 2024-09-01) is chosen as the data-set, with the first 80 % of the data-set classified as the training set and the last 20 % as the test set. The “close price” is selected as the only feature input, and a two-layer long short-term memory network (LSTM), a two-layer oneway recurrent neural network (RNN) and a one-dimensional convolutional neural network (CNN) are used to simulate the prediction of the future stock price, the accuracy of the prediction is judged by the Mean Absolute Percentage Error (MAPE). In this study, the MAPE values of three models are compared at different batch sizes. It's found that the MAPE value of the LSTM model increased by 0.35 %; MAPE value of the RNN model increased by 0.23 % and MAPE value of the CNN model increased by 1.14 % as the batch size increased. According to result, RNN is more accurate in capturing the hidden changes in stock data fluctuations and shows more accurate results in prediction.