Early Prediction of Stock Market Movements Using News Analytics and Deep Learning

Stocks are an event based on data that depend on various factors, such as a company's introduction of new products, the company's economy, and trends concerning competitors. The time series data is highly volatile, and hard to predict the nature and behaviour of the data due to its complex nature. There are many uncertainties inherent in attempting to predict stock price fluctuations. Stock prices do not follow a consistent seasonal cycle, but rather are influenced by a variety of external factors that contribute to an irregular pattern of movement over time. Predicting stock performance precisely is challenging due to the role of exogenous variables in driving market dynamics in non-linear ways. The stocks definitely have temporary and permanent rise or fall values that justifies the pulses and steps. Various research studies employ comparable metrics to predict the trajectory of the primary stock market index. These studies either use only textual data or only market data for prediction or even if they use the inter-dependency between them, the results have only been marginally better. Our proposed system uses time series stock market data and textual news data to study their correlation. We use appropriate deep-learning models and techniques to predict the possible outcome over the next ten days. A Long-Short Term Memory (LSTM) model predicts whether the stock market will increase or decrease over the next ten days. This model predicts a value between -1 and 1: -1 indicating the decrease and one indicating the increase.