Using LSTM to predict cryptocurrency prices

Inspiration

In Professor Zhang's CS206, we learned the fusion knowledge of computer science and economics. In a discussion about thinking about a research program to pursue Nobel Prize or Turing Prize, we happened to think of the interdisciplinary subject of using LLM to conduct stock trading, and were very interested in it, so we wanted to take this opportunity to put it into practice and learn related aspects.We aimed to create a tool that could analyze patterns in the price fluctuations of high-risk coins, providing investors with a clearer picture of potential future trends.

What it does

Our system utilizes LSTM to process historical data of cryptocurrencies, generating a predictive model that estimates future prices. It is convenient to use, it can automatically get the relevant currency data and use the model to train and predict future trends

How we built it

We have learned relevant knowledge through various channels. First, we read several papers that showed how well LSTM was able to predict problems that required long-term dependencies, and decided to use this approach. Then, we learned about LSTM through various platforms such as ChatGPT, youtube, bilibili, csdn and tried to build a model. Finally, we solved the problem of how to get the data, how to visualize the data and so on through various kinds of learning online. We developed our solution using Jupyter notebooks for their versatility and ease of use. We divided the data into training and testing sets to validate our model’s accuracy and using regularization to avoid overfitting.

Challenges we ran into

The main challenge was that we didn't have the knowledge of machine learning, how to build models, how to train, how to capture and visualize data. Another significant challenges was the normalization of data to prevent any model bias.

Accomplishments that we're proud of

We are particularly proud that our model was able to minimize losses after training, as can be seen from the loss graph we generated. The actual closing price and our forecast have similar trends, demonstrating the validity of our model.

What we learned

Throughout the project, we gained a deeper understanding of the key role of LSTM networks and data preprocessing related to machine learning. And learned how to visualize data, how to get data using python.

What's next for Using LSTM to predict cryptocurrency prices

Looking ahead, we plan to further refine our model’s accuracy and extend its functionality to include sentimental analysis of social medias. We also aim to broaden the scope of our model to encompass a wider range of cryptocurrencies and potentially integrate it with trading platforms for automated decision-making.