Using Unsupervised Machine Learning to examine the outcome of Cryptocurrencies data and how to analyze it.
The purpose of this project is to use Unsupervised Machine Learning to analzye a cryptocurrencies database and create a report that includes traded cryptocurrencies classified by their Unsupervised Machine Learning groups and feaures. This classification report can be used by an investment bank to propose a new cryptocurrency investment portfolio to its clients. We use the following methods for the analysis:
- Preprocessing the Data for PCA
- Reducing Data Dimensions Using PCA
- Clustering Cryptocurrencies Using K-means
- Visualizing Cryptocurrencies Results
Data Source:Crypto Data
Software: Python 3.8.8, Anconda Navigator 2.1. 1, Conda 4.11.0, Jupyter notebook 6.4.6
After preprocessing and filtering the cryptocurrency data there are a total 532 tradable cryptocurrencies.
Using Unsupervised Machine Learning there is no known output. To group the cryptocurrencies, I will be using clustering algorithms alao known as the K-Mwan and project the findings using data visualizations. Using the K-Means method the results produced the Elbow-Curve, which iterates k-values from 1-10
The best k-value appears at the Elbow-Curve of 4 and as a result an output of 4 clusters can categorize the cryptocurrencies data.
These 3D Scatter Plots were obtained using the PCA Algorithim to reduce the cryptocurrencies dimensions to three principal components
These 2D-HV Scatter plot's above were obtained using the PCA algorithm to reduce the cryptocurrency data using TotalCoinsMined and TotalCoinsSupply in the X & Y axis. The cyrptocurrency BitTorrent coin in Class #2 is the unique outlier in this HV Scatter plot.
According to the Tradable Cryptocurrency Table most of the cryptocurrency coins are from classes 0 and 1. The highlighted BitTorrent coin is the only cyrptocurrency coin in class 2.
In summary various steps such as preprocessing the data, reducing the data dimensions, using the K-means to cluster the cryptocurrencies and finally visualize the cryptocurrencies data were used to help classify the output data in the Unsupervised Machine Learning process. As a result, the data has identified the classification of 532 cryptocurrencies based on similarities of their features. This ultimately gives a wide range of investment opportunities for clients to use to diversify their portfolio investments.