BITCOIN PRICE PREDICTION USING CNN AND LSTM

click here to download project abstract

click here to download project base paper

ABSTRACT
The purpose of this study is to find out with what accuracy the direction of the price of Bitcoin can be predicted using machine learning methods. This is basically a time series prediction problem. While much research exists surrounding the use of different machine learning.

Techniques for time series prediction, research in this area relating specifically to Bitcoin is lacking. In addition, Bitcoin as a currency is in a transient stage and as a result is considerably more volatile than other currencies such as the USD. Interestingly, it is the top performing currency four out of the last five years. Thus, its prediction offers great potential and this provides motivation for research in the area. As evidenced by an analysis of the existing literature, running machine learning algorithms on a GPU as opposed to a CPU can offer significant performance improvements. This is explored by benchmarking the training of the RNN and LSTM network using both the GPU and CPU. This provides a solution to the sub research topic.

Finally, in analysing the chosen dependent variables, each variables importance is assessed using a random forest algorithm. In addition, the ability to predict the direction of the price of an asset such as Bitcoin offers the opportunity for profit to be made by trading the asset.

Keywords: Bitcoin Prediction, Time complexity, Machine-learning, Database architecture, RNN, LSTM.

INTRODUCTION

DOMAIN SPECIFIC INTRODUCTION

Time series prediction is not a new phenomenon. Prediction of most financial markets such as the stock market has been researched at large scale. Bitcoin presents an interesting parallel to this as it is a time series prediction problem in a market still in its beginning stage. As a result, there is high volatility in the market and this provides an opportunity in terms of prediction. In addition, Bitcoin is the leading cryptocurrency in the world with adoption growing consistently over time. Due to the open nature of Bitcoin it also poses another difficulty as opposed to traditional financial markets. It operates on a decentralised, peer-to-peer and trustless system in which all
transactions are posted to an open ledger called the Blockchain. This type of transparency is not seen in other financial markets. Traditional time series prediction methods such as Holt- Winters exponential smoothing models rely on linear assumptions and require data that can be broken down into trend, seasonal and noise to be effective. This type of methodology is more suitable for a task such as predicting sales where seasonal effects are present. Due to the lack of seasonality in the Bitcoin market and it’s high volatility, these methods are not very effective for this task. Given the complexity of the task, deep learning makes for an interesting technological
solution based on its performance in similar areas. Tasks such as natural language processing which are also sequential in nature and have shown promising results. This type of task uses data of a sequential nature and as a result is similar to a price prediction task. The recurrent neural network (RNN) and the long short term memory
(LSTM) flavour of artificial neural networks are favoured over the traditional multilayer perceptron (MLP) due to the temporal nature of the more advanced algorithms. The aim of this research is to ascertain with what accuracy can the price of Bitcoin be predicted using machine learning.

Title: Bitcoin Price Prediction using CNN and LSTM

Abstract:

Cryptocurrencies, particularly Bitcoin, have gained immense popularity in recent years, making their market dynamics highly volatile. Predicting Bitcoin prices is a complex task due to various influencing factors. This postgraduate student project proposes a novel approach for Bitcoin price prediction using Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. The integration of these deep learning models aims to enhance the accuracy of predictions by capturing both short-term patterns and long-term dependencies in historical Bitcoin price data.

Existing System:

The existing systems for cryptocurrency price prediction often rely on traditional statistical models, which may struggle to capture the intricate patterns present in the highly dynamic cryptocurrency market. Additionally, they may not effectively incorporate the temporal dependencies that exist in sequential data, such as historical price trends.

Proposed System:

The proposed system employs a hybrid model, combining CNN and LSTM networks to extract features from historical Bitcoin price sequences. CNNs excel at capturing local patterns, while LSTMs are proficient in modeling long-term dependencies. The fusion of these models aims to create a robust framework for accurate and reliable Bitcoin price predictions.

System Requirements:

Hardware:

• High-performance GPUs for training deep learning models efficiently.
• Sufficient RAM to handle large datasets during training.

Software:

• Python programming language for model development.
• TensorFlow or PyTorch for implementing deep learning models.
• Web framework (e.g., Flask or Django) for creating the user interface.
• Database system (e.g., MySQL or MongoDB) for storing and retrieving historical data.

Algorithms:

1. Convolutional Neural Network (CNN):

• Used for feature extraction from local patterns in historical price sequences.

1. Long Short-Term Memory (LSTM):

• Employed to capture long-term dependencies and temporal patterns in the data.

Architecture:

1. Data Collection:

• Obtain historical Bitcoin price data from reliable sources.

1. Preprocessing:

• Normalize and preprocess the data to make it suitable for training.

1. CNN Feature Extraction:

• Apply CNN to extract local features from the sequential data.

1. LSTM Modeling:

• Utilize LSTM to capture long-term dependencies and temporal patterns.

1. Hybrid Model Fusion:

• Combine CNN and LSTM outputs to create a comprehensive predictive model.

1. Training and Validation:

• Train the model using historical data and validate its performance.

1. Web User Interface:

• Develop a user-friendly web interface for users to input parameters and view predictions.

Technologies Used:

• Python
• TensorFlow or PyTorch
• Flask or Django (for web development)
• HTML, CSS, JavaScript (for web interface)
• MySQL or MongoDB (for data storage)

Web User Interface:

The web interface will include input fields for users to specify parameters such as time frame and prediction horizon. The predicted prices will be displayed graphically, providing an intuitive visualization of the model’s forecasts.

List of References:

1. Satoshi Nakamoto. (2008). “Bitcoin: A Peer-to-Peer Electronic Cash System.”
2. Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner. (1998). “Gradient-based learning applied to document recognition.”
3. S. Hochreiter and J. Schmidhuber. (1997). “Long Short-Term Memory.”
4. F. Chollet et al. (2015). “Keras: The Python Deep Learning library.”
5. A. Géron. (2019). “Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow.”

This project aims to contribute to the field of cryptocurrency price prediction by leveraging the power of deep learning models, providing a more accurate and adaptable solution for investors and researchers in the cryptocurrency space.