Machine learning has been used extensively in the financial sector for Its ability to analyze a large amount of data and extract insight. This made it especially useful to predict future trends of stocks. With the rise of blockchain and cryptocurrency, it’s no surprise that researchers have begun to look for ways to harness the power of machine learning to help investors increase their performance and enhance the trading experience [1]. For our project, we aim to create a machine learning tool that predicts the prices of various cryptocurrencies. To achieve this, we will be using a supervised learning algorithm and a dataset containing the historic prices of cryptocurrencies.
The main problem that we aim to address in this project is providing a more accurate method of predicting future cryptocurrency prices using machine learning algorithms. Since cryptocurrency is driven entirely by supply and demand and subject to less government regulation than stocks, the cryptocurrency market is a lot more volatile than the traditional stock market [1]. Therefore, we anticipate that by applying machine learning techniques on data containing the market capitalization, volume of transactions, and fear & greed index of various cryptocurrency coins, we will be able to discover trends in the cryptocurrency market and devise an accurate cryptocurrency price prediction technique [2].
One method of analyzing datasets is a random forest regression (RF). The benefit of an RF is the higher accuracy and ability to parse over large datasets. However, the algorithm tends to overfit for noisy data. Considering the volatility of cryptocurrency prices, it is possible the RF will overfit. The other method being considered is support vector machines (SVMs). SVMs can tackle both linear and nonlinear data and provide predictions based on the data. However, SVMs are not as applicable to large datasets and are more beneficial when applied to smaller datasets, such as the short-term market analysis. Further, we can study trends in cryptocurrency prices and compare them to extract and/or extrapolate long-term market trends.
The end goal of this project is to be able to predict the rise and fall of cryptocurrency prices and measure how predictable returns are for the selected cryptocurrencies. The team will aim to use two to three different methods and determine which of the methods produces more accurate results. After analyzing these results, we aim to extract general trends to have a sort of a comparative analysis between different cryptocurrencies. Due to the volatile nature of the cryptocurrency market, there is expected to be some inaccuracy in the predictions. By combining long-term and short-term data and testing different methods the algorithm will provide the most accurate results.
[1] Aharon, D. & Qadan, M., 2022. Bitcoin and the day-of-the-week effect. Science Direct. https://doi.org/10.1016/j.frl.2018.12.004.
[2] Sebastião, H., & Godinho, P. (2021, January 6). Forecasting and trading cryptocurrencies with machine learning under changing market conditions - financial innovation. Financ Innov 7, 3 (2021). https://doi.org/10.1186/s40854-020-00217-x.
[3] Weinhardt, Patrick Jaquart and David Dann and Christof. Short-term bitcoin market prediction via machine learning. https://doi.org/10.1016/j.jfds.2021.03.001.