Stock Prediction of Multivariable Using Bi-Long Short Term Memory and Capsule Neural Network

Rapid and volatile stock price movements make it difficult for people to predict a stock price, often going up or down. Two approaches are often used in analyzing stocks that is the method of fundamental and technical analysis. Both analyzes are critical to the prediction of stock price. This study proposed stock prediction using fundamental analysis and technical analysis. Fundamental analysis is based on news headlines. Technical data used opening prices, closing prices, highest prices, lowest prices, and volume. In revealing the text of fundamental parameters, where the information in the text is interrelated, it is necessary to pay attention to the relationship between words. Stock price data needs to consider the position and direction of data without eliminating the variables, time, and the relationship between time and existing variables. Therefore, fundamental data are classified through two-way Long Short Term Memory (LSTM) or BiLSTM, producing positive, neutral, or negative sentiments. The sentiment score is the output of BILSTM, and technical variables are used as a feature in predicting the next stock trend using the Capsule Neural Network (CapsNet). The sentiment analysis results using the two-way LSTM produced an accuracy of 92.03%. The stock prediction results also show that the proposed method improved by increasing the accuracy by 7% to 68.79
^%
compared to the Convolutional Neural Network.