A Survey on Stock Price Prediction Using Deep Learning

Abstract

Stock market price forecasts have been a topic that both analysts and researchers have long been interested in. Stock prices are difficult to analyse because of the excessive volatility nature that rest on many economic factors. Stock price forecasts based on historical data have proven to be inadequate. A study of sentiment analysis found a relationship between stock price movements and the publication of news articles. Many sentiment analyses use a variety of algorithms, such as support vector machines, naïve Bayesian regression, and deep learning, to look at how they are performed at different levels. The accuracy of the algorithm depends on the amount of training data provided. However, the amount of text information collected and analysed in previous studies is not yet sufficient and produces low-precision predictions.

In this paper, collect large amounts of time series data and use deep learning models to analyse related news articles to improve the accuracy of stock price forecasting. Naïve Bayesian classifiers are used to classify news texts with negative or positive emotions. Along with the number of positive and negative emotions in each day's news articles, and past data, close prices and distribution of adjacent days are used for predictive purposes and accuracy of 65.30 to 91.2% achieved in different machine learning technologies.