Authors: S.Srinivas, K. Srimukhi
Abstract: The rapid expansion of the cryptocurrency market has created significant interest in developing intelligent forecasting models capable of predicting future price movements with improved accuracy. Among various digital assets, Bitcoin exhibits substantial price volatility because of its decentralized nature, market sentiment, macroeconomic conditions, trading activity, and blockchain-related indicators. These highly dynamic characteristics make cryptocurrency price prediction a challenging task for conventional statistical techniques. This paper presents a machine learning-based framework for forecasting Bitcoin prices by integrating Random Forest Regression, Long Short-Term Memory (LSTM) networks, and a Bagging technique to capture both nonlinear relationships and temporal dependencies within historical market data. The proposed methodology utilizes a comprehensive dataset covering the period from March 31, 2015, to April 1, 2023, consisting of 47 explanatory variables categorized into multiple market-related groups. Before model development, extensive data preprocessing, feature preparation, and normalization are performed to improve learning efficiency and prediction reliability. Model performance is evaluated using standard regression metrics, including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and Directional Accuracy (DA). Experimental analysis demonstrates that the proposed machine learning framework effectively captures complex cryptocurrency market behavior and provides reliable forecasting performance for Bitcoin price prediction. The study highlights the potential of combining ensemble learning and deep learning techniques to support intelligent financial decision-making, cryptocurrency market analysis, investment planning, and risk management applications.
