Indian Stock Movement Prediction with Global Indices and Twitter Sentiment using Machine Learning

Abstract

In recent times, there is great interest shown in the stock market activities, for reasons like unpredictability of circumstances due to the pandemic situation. Since stock market procedures are extremely dynamic in nature and it is very challenging to do any kind of prediction, employing Machine Learning algorithms to do so is but natural. We are interested particularly in exploring the situation in Indian Stock Market. In this paper, we describe the implementation of the Long Short-Term Memory (LSTM) network, Gated Recurrent Unit (GRU) networks for stock prediction. The prediction is performed for the closing prices of stocks of twenty-five Indian companies. The results indicate that a two-layer GRU outperforms all other networks as far as these twenty-five companies are concerned. We have predicted the stock market opening of next day using the closing of global market indices, concluding that there is a high correlation between the global and Indian market movement. Work on how the twitter financial sentiment effects the stock market has been performed by predicting the change in price over the week using twitter sentiment. The tweets were divided into three categories, positive, negative, and neutral. We have used support vector machine (SVM), Gradient boost and XGBoost, of which Gradient Boost provided the best results. The accuracies of the methods we have implemented for all the three tasks-predicting stock opening price, using historic data and global indices; range between a good 93% to 99%. In case of prediction using twitter sentiment, it ranges from 85% to 91 % when relevant financial tweets are available. The work has a natural extension to study robustness of our model for the pandemic year 2020-2021; which is currently under progress.