Stock Price Prediction Using a CNN-LSTM Based Deep Learning Approach

Abstract

Early and precise stock price forecasting has become a crucial aspect of financial risk reduction and a better approach towards investing in the past few years as a result of the volatility in financial markets. The conventional analytical methods are based on manual interpretation and statistical measures, which can be incapable of quantifying nonlinear relationships existing in the market data. To overcome these drawbacks, we have come up with a hybrid deep learning model of stock price prediction, which is a hybrid of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. Technical indicators and subjective estimates are usually useful in the stock price prediction and may be subjective as well as lengthy across various analysts. The data has stock data (Open, High, Low, Close and Volume) of some of the most successful technology companies such as AAPL, AMD, IBM, GOOGL, AMZN, NVDA, EBAY and CSCO on a daily basis. The data preprocessing will involve data cleaning, 20 days moving average (MA20) and Min-Max normalization to give a stabilized training behaviour. These representations are then inputted to LSTM layers which are used to model long term dependencies and sequence relations of price changes of stocks. This stratification ensures that reuse of features is also efficient not to mention that the valuable time information is saved. The network is computationally efficient and can be trained without the use of a lot of hardware hence can be applied in practice. This model was trained and tested on 2019-2024 data related to the stock market. Experimental outcomes show that it has better predictive performance than compared to traditional standalone models with reduced Mean Squared Error (MSE). In order to increase usability, an interactive analysis interface was created with the help of Plotly that allows visualizing the comparison between the predicted and the actual stock prices. The proposed framework offers the solution of financial time-series forecasting based on a scalable and efficient approach and can help investors and analysts to make well-informed decisions. The system is flexible to be incorporated in web-based financial analysis systems and it can facilitate on-the-fly market analysis.