Application of machine learning techniques on cryptocurrency index prediction
Cryptocurrencies are digital currencies which all transactions are verified, and their records maintained in an immutable ledger by a decentralized system that uses cryptography. In the past 10 years, the cryptocurrency market has been used in business and in the financial market with its market cap...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Final Year Project |
Language: | English |
Published: |
Nanyang Technological University
2021
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/153207 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | Cryptocurrencies are digital currencies which all transactions are verified, and their records maintained in an immutable ledger by a decentralized system that uses cryptography. In the past 10 years, the cryptocurrency market has been used in business and in the financial market with its market capitalization exceeding 2 trillion dollars in 2021. Thus, it has become an asset class that has attracted massive media attention and interested investors worldwide. However, due to its nature of massive price fluctuations and multiple flash crashes, it has been a challenge to predict their prices. Additionally, due to its infancy, cryptocurrency price prediction has been focused purely on technical indicators and of late included analysis of social websites as well. This project aims to integrate and evaluate modern machine learning models to predict the closing price of a cryptocurrency index. This project uses the CCi30 index, a collective, rule-based index for cryptocurrencies. Technical features are derived from the index and content-based features are specifically gathered and analysed from specific accounts from social media site Twitter. The combination of these features was fed as inputs into various machine learning models to predict the price action of CCi30. This project utilizes machine learning models such as XGBoost (a gradient boosting tree), long short-term memory (LSTM) (an artificial recurrent Neural Network (RNN)) and KNearest-Neighbours (KNN) (a supervised classifier). All the models are tuned with Grid Search and are evaluated using standard regression metrics. Text sentiment analysis is applied using a lexicon rule-based model that is used to evaluate and understand texts from social media tweets. |
---|