Lazáro Javier Hernández, an engineer and researcher at the Universidad Internacional Iberoamericana (International Ibero-American University, UNIB), is participating, alongside an international team, in a study that compares different artificial intelligence models to improve short-term cryptocurrency price predictions. The research also analyzes whether incorporating information about user reactions on X (formerly Twitter) can provide useful data for making these predictions.
Cryptocurrency prices can change rapidly and be affected by multiple factors. This variability makes anticipating their behavior—even over short periods—a challenge for researchers.
Given this challenge, the study, titled “Cryptocurrency Market Trends: A Machine Learning-Driven Time Series Forecasting with Twitter Sentiment Integration,” proposes an AI-based strategy that combines different methods of data analysis. Specifically, it compares three hybrid models: VAR-LSTM, XGBoost-LSTM, and CNN-LSTM. The goal is to determine which model yields the best results when attempting to predict the closing price of Bitcoin, Ethereum, and Dogecoin for the next hour.
Unlike models that use a single technique, these alternatives combine different tools to analyze both available market information and price trends over time.
Three Models to Find a Better Prediction
To conduct the comparison, the researchers used historical cryptocurrency data, such as opening and closing prices, high and low values, and trading volume. They also incorporated various indicators that help identify trends and changes in price behavior.
All three models were subjected to the same evaluation conditions. This allowed the researchers to determine which model produced the most accurate predictions and whether any offered advantages over the others.
Among the three alternatives, XGBoost-LSTM performed best, with 80% accuracy in predicting whether the price would rise or fall over the next hour. CNN-LSTM achieved 67.2%, while VAR-LSTM achieved 59.4%.
What Can User Reactions on X Contribute?
In addition to comparing the three models, the study analyzed whether user reactions on X could provide information to help improve predictions. To do this, the researchers incorporated information about the sentiment of posts related to Bitcoin, Ethereum, and Dogecoin into the analysis. The posts were classified based on whether they expressed a positive, negative, or neutral reaction, and this data was incorporated alongside market information.
The results showed that this information had a significant effect. When the researchers removed user reactions from the XGBoost-LSTM model, its accuracy in predicting price direction dropped from 80% to 67%. The difference was statistically significant.
This result suggests that real-time information from social media can provide additional signals to historical market data, although it does not replace such data or eliminate the uncertainty inherent in cryptocurrencies.
An Approach with Potential for Further Development
The study shows that combining different artificial intelligence models can yield better results than using a single technique to predict short-term price movements. At the same time, the research highlights the potential value of incorporating external information—such as user reactions expressed on X—into historical price behavior.
However, the researchers also acknowledge certain limitations, such as the use of a single social media platform and the possibility that the analyzed posts may include automated activity or certain biases. They also note that the data used corresponds to a period prior to several significant changes that the cryptocurrency market subsequently underwent.
As areas for future research, the study proposes exploring new artificial intelligence models, incorporating information from different social media platforms, and adding other data related to cryptocurrency network activity.
If you’d like to learn more about this study, click here.
To read more research, check out the UNIB repository.
Artificial Intelligence is changing the way information is analyzed; professionals capable of leading these technologies are needed. In this context, UNIB offers the Master’s in Strategic Management with a Specialization in Information Technology, designed for professionals seeking to develop the skills to lead technology and organizational transformation projects. The program can be pursued with a scholarship provided by the Ibero-American University Foundation (FUNIBER). Learn more about the program and request information about your scholarship.