Cryptocurrency Clustering Analysis using K-Means

This project explores the impact of using fewer features to cluster cryptocurrency data using the K-Means algorithm. The clustering was performed using both the original feature set and a reduced feature set (via PCA or manual selection), followed by visual analysis and comparison of the results.

---

Objective

To evaluate how reducing the number of features affects the performance and visual clarity of clusters generated through K-Means clustering.

---

Dataset

The dataset contains various time-based percentage price changes for multiple cryptocurrencies, including:

• 24h
• 7d
• 14d
• 30d
• 60d
• 200d
• 1y

---

1. Elbow Curve Comparison

Elbow Curve 1 (Original Data):

• Built using all 7 original time-based features.
• Shows a gradual decrease in inertia without a sharp elbow.
• Higher overall inertia values indicate looser clusters and possibly noisy data due to high dimensionality.

Elbow Curve 2 (Reduced Features via PCA or Selection):

• Demonstrates a clear elbow point and lower inertia values.
• Suggests better-defined clusters and tighter groupings of data points.
• Indicates improved clustering efficiency by removing redundant or less informative features.

---

2. Cluster Scatter Plot Comparison

Cluster Graph 1 (Original Data):

• Clusters appear overlapping, with no distinct boundaries.
• Difficult to interpret due to clutter caused by high-dimensional input.
• Indicates that clustering with all features may not effectively reveal meaningful groupings.

Cluster Graph 2 (Reduced Features):

• Shows clear separation between clusters.
• Easily identifiable and visually distinct groupings (e.g., 3–4 clusters).
• Indicates that reducing the number of features helps highlight key patterns and removes noise.

---

Conclusion

Using fewer features (via dimensionality reduction or careful selection) for K-Means clustering of cryptocurrency data results in:

• More compact and separable clusters.
• Improved visual clarity and interpretability.
• Reduced noise and redundancy, allowing for better algorithm performance.

Fewer, more informative features = more meaningful clustering results.

---

📂 Files

• crypto_market_data.csv – Raw dataset of cryptocurrency price changes.
• Crypto_Clustering.ipynb – Jupyter Notebook for preprocessing, clustering, and visualization.
• README.md – Summary and visual analysis of clustering results.

---

Tools Used

• Python
• scikit-learn
• pandas
• hvPlot
• PCA (Principal Component Analysis)
• K-Means Clustering

---