mznaturl1/cryptoclustering

CryptoClustering

The requirements for this challenge were to use unsupervised learning technique of k-Means clustering to group cryptocurrencies by their performance to create portfolio portfolio recommendations.

Data Used

crypto_market_data.csv - market data of different cryptocurrencies during different time periods

Summary

Using the elbow curve method to normalize the data to find the optimal k value for the k-Means model that will use all of the original features of the dataset.

Elbow curve plot showing a value of 4 for k to be optimal for the dataset with all features

A k-Means model was trained and predicted using the best k values, resulting in four clusters of cryptocurrencies. The inertia of each cluster was large enough to consider reducing the number of features.

A scatter plot showing 4 clusters with heavy inertia

To reduce the amount of features used, the Principal Component Analysis (PCA) was applied to create three primary clusters.

DataFrame holding 3 primary clusters as columns and cryptocurrency as inde

Then the PCA data was used to recalculate the optimal k value for the k-Means model.

Elbow curve line plot from the PCA data that shows 4 to be the optimal k value

Finally, a new cluster was drawn using the best k value of the PCA feature.

Scatter plot showing 4 low inertia clusters generated using the PCA dataframe

---

Technologies

This project uses Jupyter Notebook using a Python 3 kernel.

Dependencies used:

1. [Jupyter] - Running code
2. [Conda] - Dev environment
3. [Pandas] - Data analysis
4. [Matplotlib] - Data visualization
5. [Numpy] - Data calculations & Pandas support
6. [hvPlot] - Interactive Pandas plots
7. [scikit-learn] - kMeans clustering, PCA, and StandardScaler

---