Robust and Predictive Deep Neural Networks for Stock Price Forecasting

Forecasting stock prices has become a widely researched domain, and conventional prediction methods primarily rely on statistical and econometric models. Nevertheless, these methods encounter challenges when confronted with dynamic, nonstationary time series data. The advent of the internet alongside the surging prevalence of social media has brought about a significant shift. Online news snippets and user comments now frequently mirror investor sentiments and perspectives on various stocks, encompassing a wealth of crucial insights that can greatly enhance stock price prediction. In this paper, forecasting of stock prices has been implemented using two cutting-edge deep learning models, namely Long Short-Term Memory and Gated Recurrent Unit. Both models have been evaluated on two publicly available datasets, namely the Google stock price dataset and Tesla stock price dataset. Experimental results show that LSTM has achieved accuracies of 87.91% and 85.67%, while GRU has achieved accuracies of 93.62% and 88.78% on the Google dataset and Tesla dataset, respectively. Keeping in view the complexity of the dataset, the obtained results, in turn, pave the way for practical applications within the dynamic realm of real-time stock market prediction.