Predicting Stock Price Movements Using Lstm Networks

Abstract

This research explored a deep learning approach using a Bi-Directional LSTM network to forecast stock market prices. The proposed model utilizes historical stock data as input to generate future price estimates. A prediction window of 30 days was applied, and the model was trained and evaluated on datasets from the New York Stock Exchange (NYSE), the Nikkei 225, and the Nasdaq Composite. The method achieved a Mean Absolute Percentage Error (MAPE) of 0.014 for the NYSE, 0.01 for the Nikkei 225, and 0.018 for the Nasdaq Composite. When compared with results reported in the reference study, the model demonstrated a notable enhancement in prediction accuracy. For future research, the framework may be extended to other global stock exchanges and enriched with additional factors, such as news sentiment analysis, to further boost performance. Overall, the findings highlight the potential of this method in advancing stock price prediction, offering valuable insights for investors and financial analysts in decision-making.