Incorporating Financial News Sentiments and MLP-Regressor with Feed-Forward for Stock Market Prediction

EasyChair Preprint 5832

Abstract

Stock Market being very volatile depends on various political, environmental, and internal factors. The stock prices prediction using news data is an interesting research topic. In this paper, an approach is proposed that represents textual news data as sentiment metrics using VADER sentiment analysis and price data scaled down between 0 and 1. The output predicted price of a stock on a particular day is fed forward to the next level of MLP*-Regressor to train as well predict the prices of following days. Experiments have been conducted on 10-year financial news as well price data of Reliance Company using the proposed model. The results show that the model because of feed-forward was able to learn the trend and the depths were followed more closely than the spikes. The model was tested on the news data of the same date as well as on the previous date separately. The model is an improvement made to MLP-Regressor whose results are also compared. The MLP-Regressor with feed-forward was able to learn long-term trends and also predict with an accuracy of .714 for the upcoming 7 days.

*MLP- Multilevel Perceptron

Keyphrases: Forecasting, MLP Regressor, News sentiment analysis, Stock Market Prediction, Stock Market Price, Stock Prediction, Time series data, stock market, time series

@booklet{EasyChair:5832,
  author    = {Junaid Maqbool and Preeti Aggarwal and Ravreet Kaur},
  title     = {Incorporating Financial News Sentiments and MLP-Regressor with Feed-Forward for Stock Market Prediction},
  howpublished = {EasyChair Preprint 5832},
  year      = {EasyChair, 2021}}