BullBear AI Pro: A Hybrid Real-Time Framework for Stock and Cryptocurrency Trend Prediction Using Random Forest and Long Short-Term Memory Networks

Abstract

Financial markets are highly dynamic, nonlinear, and influenced by a wide range of economic, political, and behavioral factors, making short-term trend prediction one of the most challenging problems in computational finance. The continuous interaction of macroeconomic indicators, company-specific developments, investor sentiment, and unexpected global events produces noisy and non-stationary time series that are difficult to model using conventional statistical techniques. Although traditional forecasting methods and rule-based technical analysis remain widely used, they often fail to capture complex temporal dependencies and nonlinear relationships present in stock and cryptocurrency price movements. In recent years, machine learning and deep learning approaches have demonstrated significant potential for extracting predictive patterns directly from historical market data and supporting more informed investment decisions. This paper presents BullBear AI Pro, a real-time hybrid forecasting framework that combines a Random Forest classifier and a Long Short-Term Memory (LSTM) network to predict next-period market direction (UP or DOWN). Historical market data are automatically retrieved from Yahoo Finance and processed through a feature engineering pipeline that computes widely used technical indicators, including moving averages, Relative Strength Index (RSI), Moving Average Convergence Divergence (MACD), and Bollinger Bands. Predictions generated by both models are integrated using an ensemble decision mechanism that produces confidence-based BUY, SELL, and HOLD recommendations. The framework also includes an interactive dashboard developed using Streamlit, enabling users to train models, monitor multiple stocks and cryptocurrencies, and obtain live predictions in real time. Experimental evaluation was conducted using approximately ten years of historical data for Apple, Amazon, Microsoft, and Tesla. The results show that the LSTM model consistently outperformed the Random Forest classifier, achieving an average accuracy of 53.43% and an average F1-score of 69.63%, compared with 48.42% accuracy and 36.41% F1-score for the Random Forest model. These findings demonstrate that temporal deep learning models are more effective in capturing market dynamics, while the proposed hybrid framework provides a practical, scalable, and deployable decision support system for real-time stock and cryptocurrency forecasting.