ABSTRACT
In this study, we develop a hybrid predictive framework to forecast Bitcoin prices using both deep learning and traditional machine learning approaches. The dataset consists of historical Bitcoin trading data enriched with engineered features such as moving averages, Relative Strength Index (RSI), and time-based variables. A Long Short-Term Memory (LSTM) network is trained to perform multi-output regression to simultaneously predict five key market indicators: Open, High, Low, Close prices, and trading Volume. To provide a baseline for performance comparison, a MultiOutput Linear Regression model is also implemented. The models are evaluated using Root Mean Squared Error (RMSE) and the coefficient of determination (R²) for each target variable. The LSTM model demonstrates superior predictive performance, particularly in capturing temporal dependencies inherent in financial time series. Finally, the models are saved for future deployment, and visualizations are presented to highlight the accuracy of Close price predictions. This work contributes toward the development of intelligent trading systems capable of real-time cryptocurrency forecasting.

Keywords: Bitcoin prediction, time series forecasting, LSTM, multi-output regression, linear regression, cryptocurrency, deep learning, technical indicators, RSI, moving average, RMSE, R² score.


OBJECTIVE
The primary objective of this project is to develop and evaluate a predictive framework for forecasting key Bitcoin market indicators using both deep learning and traditional machine learning techniques. Specifically, the goals of the study are as follows:
1. To preprocess and engineer meaningful features from raw Bitcoin trading data, including technical indicators (e.g., Moving Averages and RSI) and time-based features (e.g., hour, day, and month), in order to enhance the predictive capacity of the models.
2. To build and train a Long Short-Term Memory (LSTM) neural network capable of performing multi-output regression for simultaneously forecasting the Open, High, Low, Close prices, and trading Volume of Bitcoin.
3. To implement a MultiOutput Linear Regression model as a baseline to compare the performance of the deep learning approach with a traditional machine learning method.
4. To evaluate and compare the models using robust statistical metrics such as Root Mean Squared Error (RMSE) and the coefficient of determination (R² score), in order to assess predictive accuracy and model generalizability.
5. To visualize prediction results, particularly for the Close price, to better understand how each model performs in capturing temporal trends in Bitcoin price movement.
6. To save the trained models in a reusable format, enabling deployment in real-world applications such as algorithmic trading systems or financial analytics dashboards.
Through these objectives, the project aims to demonstrate the feasibility and effectiveness of using data-driven approaches for short-term cryptocurrency price prediction, ultimately contributing to intelligent financial decision-making in volatile markets.