FORECASTING EQUITY MARKET VALUES USING CONVOLUTIONAL AND LONG SHORT-TERM MEMORY NETWORKS

Abstract

This experiment explores the effectiveness of combining Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) in forecasting stock prices, addressing the challenging nature of stock market prediction. Conventional methods often struggle to capture the intricate patterns within financial market data, but deep learning techniques, such as CNNs, offer promising opportunities for improved accuracy. The study employs historical stock price data for Microsoft Corporation (MSFT) from January 1, 2013, to May 18, 2018, sourced from the Quandl API. Data preprocessing involves standardizing high, low, open, and close prices and creating input sequences representing six days of stock price data to capture temporal dependencies. The CNN architecture incorporates 1D convolutional layers, max pooling, dropout regularization, and a final dense layer for prediction. Training employs the mean squared error (MSE) loss function and the Adam optimizer. The dataset is divided into 80% for training and 20% for testing. By optimizing CNNs for stock chart images using techniques like residual learning and a bottleneck architecture, this research seeks to reveal hidden patterns within the data, ultimately contributing to the enhancement of stock price prediction accuracy.