Dynamic Sliding Window and Neighborhood LSTM-Based Model for Stock Price Prediction

Stock price prediction is a crucial task in the decision-making of investors. Forecasting successfully on future stock prices will help the investors to gain profit. However, it is difficult to predict exactly the trend of the stock market due to the complex relationship between stock prices and external factors such as news, global economy, public sentiments, and other sensitive financial information. Therefore, historical prices are the main investigation data for researchers since collecting external factors is still a challenge. Our motivation comes from the stock price’s varying trend and the relationship among stocks. To this end, we propose two approaches based on the long short-term memory (LSTM) model along with a deep concern in the historical data of stocks. (i) The input data of the LSTM model are enriched by integrating the historical price of stocks with their nearest neighbors in terms of similarity; (ii) a dynamic sliding window is applied in the prediction phase of the LSTM model based on the analysis of the significant change of the stock price. Experimental results on three stock datasets of the United States, Germany, and Vietnam show that our model outperforms the competitors in predicting the trend of stock price fluctuations. This paper is an extended version of our previous work at the IEEE International Conference on Big Data and Smart Computing (BigComp 2021). © 2022, The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd.