Unicsoft

AI model for analyzing the cryptocurrency market

Challenge

Crypto industry reached the peak of its popularity last year. There was a need to create useful tools for crypto analysis.

The client came to us with the startup idea to develop an AI model for analyzing the cryptocurrency market. As he has already built an algorithm for cryptocurrency auto-trading, he decided to add cryptocurrency price prediction feature to it as well.

The solution combines natural language processing, sentiment analysis, and machine learning algorithms (LSTM) to predict prices for cryptocurrency. It crunches huge amounts of data and tells you which cryptocurrencies to invest in and attempts to maximize your returns.

The product brings value to:

• Private traders who want to receive hints from the system of artificial intelligence and a forecast on the trend of cryptocurrency prices;
• Professional companies that need to evaluate a comprehensive forecast for the whole real-time cryptocurrency market, portfolio management, investments, risk management, etc.

Solution

Unicsoft dedicated team was in charge of developing a Cryptocurrency price prediction solution based on LSTM models and sentiment analysis of Crypto forums and channels. The model provides the prediction of 4 type of price: ‘Open’, ‘High’, ‘Low’, ‘Close’ for every Cryptocurrency individually.

The processes of price prediction model is recomputed constantly by AI Engine, potentially in a constant update loop. The model works in real time.

The model accepts any number of input parameters (the network architecture allows you to add many technical market characteristics of a selected cryptocurrency), and gives a forecast at specified prices (‘Open’, ‘High’, ‘Low’, or ‘Closed’ ) at the output. The architecture also provides a simple way to add new parameters to the line of the forecast in the future: for example, the forecast of the Volume of funds for this cryptocurrency or another indicator. No significant code changes will be required: flexible parameters are already embedded in the solution.

At present, the models work with a time step of one day, and make forecasts for 7 days ahead (forecast interval is adjustable); however, the solution can be applied to hourly input data and the forecast horizon set in hours.

The solution was deployed in Python, Jupyter Notebook deployed for demo purposes. All documentation and the code was delivered to the customer on time and according to the initial requirements.

Result

The team created the fully ready working solution for Cryptocurrency price prediction within four weeks.

The main challenges were:

• Research and selection of the optimal model architecture for Cryptocurrency price prediction within this solution.
• Improving prediction quality, optimizing time and resources for model training and real-time prediction. In the result the client received a model with 73% – 90% accuracy of prediction.

Also, we provided the client with the demonstration materials about this solution. It’s allowed the client to demonstrate the capabilities of this solution both to specialists in AI and to investment market traders as well as to those interested in investing in this project.  So the Client has an opportunity to attract new users and investors to his AI based product.