Stock Prices Prediction Based on Social Influence & Historic Data

Stock market price prediction is a challenging issue as a range of elements including political statements, economic circumstances, business market value, historical stock price, and so on, influences it. Hence study exhibit that many prebuild models like ARIMA and deep learning model like LSTM are developed but their efficiency is not up to mark for stock price prediction. In this paper, we build hybrid model which is blended with CNN and LSTM to improve the performance. We used historical data (prior stock price) in the form of numerical information from of the NIFTY50 from 2015 to 2020, as well as news data in textual form from the @NDTVProfit twitter account. In addition, we used a variety of prebuild models and deep learning models to forecast the next 10 days' values. We initiated with numerals/historical dataset and applied ARIMA, SARIMAX, Facebook prophet and LSTM on historical datasets and obtained error score 1062,964,709 and 285 respectively. In addition, models as ARIMA, SARIMAX, Facebook prophet and LSTM have been applied on combined dataset (historical datasets and news datasets) and obtained error score 789,655,380 and 170. The new hybrid model, which is blended with CNN, and LSTM deep learning models is applied on combined dataset and 89 error score was obtained which is better as compared to all previous models.