Time Series Forecasting of Stock Prices Using Neural Networks LSTM and GAN

The project seeks to improve the precision of Amazon stock price prediction using advanced neural network techniques, specifically Long Short-Term Memory (LSTM) networks and Generative Adversarial Networks (GANs). The analysis is based on OHLCV (Open, High, Low, Close, Volume) data, and various models were employed, including both conventional statistical methods like ARIMA and Fourier Transform models, as well as cutting-edge deep learning approaches. Feature extraction plays a critical role in improving model performance, with 19 features being created, such as momentum, Bollinger Bands, moving averages, and exponential moving averages (EMA), which significantly improved the accuracy of predictions. While ARIMA and Fourier Transform models yielded precise results with low AIC, BIC values, and close p-values, LSTM models faced challenges in short-term price prediction. Despite utilizing both single and multiple features, LSTMs were more effective in forecasting long-term trends rather than capturing short-term price fluctuations. Sentiment analysis from news and Reddit data is also incorporated to further refine stock price predictions. In addition to LSTMs, GANs were utilized to generate synthetic stock data for further training the models, where the historical trends are created using a Generator, and CNN acts as the Discriminator. By clubbing the advantages of neural networks for processing sequential data and also the use of GANs for generating realistic stock data, this unique method aims to increase the overall model accuracy.