ENHANCING STOCK MARKET PREDICTION WITH SENTIMENT ANALYSIS AND DEEP LEARNING

Abstract

In this report, there is a discussion on novel technique for stock price prediction using an innovative model by combining both Variational Autoencoders (VAE) and Generative Adversarial Networks (GAN) along with historical price data and sentiment analysis. Old traditional methods used for stock price prediction always have some and there is a complexity related to financial time series due to their non-linear nature. To handle these issues, our methodology uses a VAE to generate robust embeddings that fuses sentiments and historical price to find meaningful information. GAN architecture consists of GRU-based generator and CNN-based discriminator. The Generator generate realistic data and try to replicate the temporal dependency and stock time series patterns which observed in real stock price movements. The Discriminator find these created samples and try to discriminate between real and generated sample by using an adversarial training process. This competitive interaction between generator and discriminator allows the model to continuously improve which result in a better accuracy GAN for stock price predictions. The proposed model shows its capability by outperforming the previous traditional models like LSTM, GRU & GAN alone by effectively combining historical price and sentiment analysis using VAE and GAN. This report provides a detailed discussion into the architecture, implementation, and performance of the proposed model and highlight its potential to be used as an efficient tool for financial market analysis and prediction.