AN ENSEMBLE MODELLING APPROACH FOR STOCK MARKET TREND PREDICTION USING SENTIMENT ANALYSIS

Abstract

The swift advancement of the economy prompts individuals to increasingly invest in the stock market; yet, many stay reluctant owing to uncertainties regarding next-day profits.  Due to the volatility of stock markets caused by many factors, such as demonetization, next-day predictions are crucial for investors.  Stock market prediction mostly relies on a company's historical data, and many studies have utilized deep learning for stock market forecasting, with models demonstrating enhanced performance.  Nonetheless, deep learning models require a large set of input variables to enhance performance. As the number of variables increases, the quantity of parameters also escalates, which leads to the issue of overfitting.  In this research, LSTM and RF are merged with SA to resolve this issue.  Four worldwide stock indices—S&P 500, KOSPI 200, SSE, and KSE 100—are used with 43 technical indicators for the experimental validation of the selected models.  The RMSE of the proposed SMTP model for the S&P500, KOSPI200, SSE, and KSE100 are 16.9, 5.7, 44.11, and 33.19, respectively, whereas the forecast accuracies for these indices are 71.21, 95.64, 83.22, and 98.86.  This demonstrates a significant level of accuracy, indicating that the model can facilitate educated investment decisions and serve as a valuable resource for investors