Teresius AI – Predicting Cryptocurrency Trends with Deep Learning

Teresius AI Bot

About the Project

The TERESIUS_AI_BOT application (@teresius_ai_bot) predicts cryptocurrency prices using a graphical format across multiple time intervals: 30 minutes, 1 hour, 2 hours, 4 hours, 1 day, and 1 week (30m, 1h, 2h, 4h, 1d, 1w). Currently, the model supports four cryptocurrency pairs: BTCUSDT, ETHUSDT, BCHUSDT, and BNBUSDT. The number of supported pairs will expand as demand increases.

This application is built entirely on deep learning neural networks. Unlike most cryptocurrency and Forex forecasting systems, Teresius AI provides specific future price estimates for high, low, and close values, along with an associated probability of error. These errors vary over time based on market conditions and are visualized in forecast graphs (Figure 1) as root mean square errors (RMSE). The system also generates local forecast RMSE values and root mean square volatility of market parameters, which adjust dynamically with market fluctuations.

Empirical testing with various trading algorithms suggests that these forecasts can support effective trading strategies. The pricing data and models are trained on historical Binance futures market data, but they are applicable to spot markets and the same trading pairs on other exchanges.

The application is publicly accessible without registration. A brief system overview and user guide can be found in the TERESIUS_FORE section on Telegram - @teresius_ai_bot.

Principles of Prediction

The neural network models are trained on historical price data, including open, high, low, close, and volume, from the past five years of futures market activity. The dataset used for training incorporates advanced analytical techniques to identify market patterns and apply multifractal analysis to enhance prediction accuracy. The system also leverages pattern similarities across different timeframes, refining forecasts for each specific interval.

Forecast Accuracy

As of February 25, 2025, the forecast error for BTCUSDT close prices averaged 127.5 USDT, based on retrospective data. Meanwhile, the root mean square volatility of the close price over the last 80 hours was 804.1 USDT.

In Fig 1. Teresius AI Bot creates a forecast for BTCUSDT for the 4-hour interval.

(Figure 1: Forecast graph for BTCUSDT – 4h timeframe) (Telegram bot: @teresius_ai_bot)

Continuous Model Improvement

The models are continuously retrained and refined as new market data becomes available. The number of supported cryptocurrencies is not restricted, except for limitations related to the minimum historical data required for model training. However, the multifractal approach used in training allows the system to overcome many of these limitations.

Application to Forex Market Predictions

The forecasting technology is applicable to the Forex market. However, in this version, Forex predictions are not yet included.

Where to Find

Try Teresius AI Bot today at Telegram: @teresius_ai_bot

Teresius AI Bot is available on Telegram for public testing.

Share Feedback

Email us at hi@pieoneers.com, add “Teresius” in the subject.