Improved LSTM-Based Stock Prediction Method

Improved LSTM-Based Stock Prediction Method

In response to the common lag issue of Long Short-Term Memory (LSTM) networks in stock prediction, an improved LSTM-based stock prediction method is proposed. First, by using multi-dimensional vector input, daily closing prices of other companies with a high correlation to stock prices are selected, combined with the stock’s own price data as the input vector for the model; secondly, different feature vectors are selected as input vectors through feature engineering, and a combination of feature vectors that significantly reduces prediction lag is obtained through repeated training; finally, sentiment analysis of news texts related to the stock company is performed, and the resulting sentiment scores are used as input vectors for the model. The prediction results of Tencent’s stock indicate that this method not only improves prediction accuracy but also significantly mitigates prediction lag.