CryptoCastAI: A Unified AI‐Driven System for Cryptocurrency Market Forecasting and Model Comparison

Objective: The markets of cryptocurrencies are highly volatile, and the trading is constant, which makes it especially difficult to predict the prices correctly. The majority of current prediction methods are mostly based on historical data and do not remain applicable in real‐time. In this work, the authors provide a proposal of CryptoCastAI , a two‐pipeline artificial intelligence‐based cryptocurrency forecasting model, which can be used to complement the practical application and decision support by incorporating both the analysis of the historical data and the analysis of the market (in real time). Method: The suggested framework includes the use of three complementary forecasting models: Prophet that allows modelling interpretable trends and seasonality, XGBoost that allows modelling nonlinear price dynamics, and Long Short‐Term Memory (LSTM) networks that allow sequential learning of temporal patterns. Historical price data of Bitcoin and Ethereum were used to analyse the model performance, and the performance of real‐time data was evaluated using the live market data, which were obtained via the CoinGecko Application Programming Interface (API). Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE) and coefficient of determination ( R 2 ) were used to assess model performance. The system was implemented through an interactive system in Streamlit that allowed visualising predictions, volatility, and performance of the models in real‐time. Result: The experimental findings on the static historical data show that XGBoost was the best model with an MAE of 140.72, RMSE of 266.52, MAPE of 7.30%, and an R 2 of 0.46, which has a higher predictive accuracy than other models. Prophet had a good trend and seasonality interpretation but lower predictive performance (MAE = 243.75, RMSE = 415.26, R 2 = −0.49). LSTM performed the worst in the static environment (MAE = 363.39, RMSE = 443.70, R 2 = −0.12) and had weaknesses in forecasting real‐time as it is sensitive to short‐term market fluctuations. Conclusion: The results indicate that the combination of multiple forecasting machine learning algorithms in a single, real‐time framework can improve predictive power as well as interpretability in cryptocurrency price forecasting. The CryptoCastAI is a good solution to the problem of implementing academic forecasting models in the real world by providing a combination of live data ingestion, model comparison, and user‐friendly visualisation. The suggested system can be a scalable and interpretable decision‐support system to analyse cryptocurrency markets in real‐time and emphasise the feasible benefits of ensemble‐based AI forecasting strategies.