Performance Comparison of LSTM and XGBOOST for Ether Price Prediction from Spam Filtered Tweets

This study compares the Extreme Gradient Boosting (XGBoost) and Long Short-Term Memory (LSTM) models for predicting the price of the cryptocurrency Ether from processed and spam-filtered Twitter data. Ether, the cryptocurrency of the Ethereum network, has been identified as a potential asset after Bitcoin. Social media, particularly Twitter, is a major contributor to Ether's price fluctuation. Twitter is a main source of information for many small-time investors because it is one of the most prominent social media venues for interactions and news. XGBoost has shown significant potential in tackling regression issues with limited datasets, such as weather and demand prediction. LSTM, on the other hand, is widely recognized as one of the best and most simple deep learning algorithms for forecasting problems. In this study, we examine and compare the LSTM model's base version with the XGBoost model for predicting Ether prices. These price forecasts are supported by sentiment analysis performed on Twitter data in the form of tweets, which are spam filtered using the C5.0 algorithm to decrease noise in the data. This work advances the subject of price prediction by comparing two models using the same dataset and computational environment.