Forecasting Stock Prices using Tweets

Stock market price prediction is a challenging problem since the market is an immensely complex, stochastic and dynamic environment. There are many studies from various areas aiming to improve the performance of prediction and analysis of public emotion has been the focus of one of them. We use information shared over Kaggle, an online community of data scientists and machine learning practitioners, to better understand and predict stock prices of Tesla. This article studies the methods to preprocess tweets and to tune models so that neural network models and linear regression can adapt to the preprocessed tweets. According to previous authors, one way to preprocess tweets is to keep the tweets of smart user, and output can be the next-day close prices or the next-day return. For that goal, prediction models (CNN-LSTM, LSTM and linear regression) were built and modified step by step and their results were analyzed by Mean Square Error and Mean Absolute Error. Finally, the LSTM model, with close value and weighted labels as input features and return as output feature, wins the prediction of stock price of Tesla among other candidate models with different input and output features.