This repository features an advanced Unsupervised Machine Learning benchmarking pipeline that evaluates and compares three fundamentally different clustering paradigms: Centroid-based (K-Means), Connectivity-based (Agglomerative Hierarchical), and Density-based (HDBSCAN) clustering.
The project demonstrates how mathematical variations in algorithmic design influence cluster boundary detection, handling of geometric density, and noise isolation.
- Multi-Paradigm Benchmarking: Implemented and optimized three distinct clustering frameworks to observe variance minimization vs. spatial density trajectory mapping.
- Mathematical Scaling: Deployed StandardScaler pipelines to neutralize directional scale bias across botanical features (
sepalandpetaldimensions). - Noise & Outlier Engineering: Leveraged HDBSCAN's internal lambda-density hierarchies to actively isolate transitional overlapping data points as statistical noise without distorting core cluster profiles.
- Production-Ready Architecture: Structured preprocessing workflows into reusable, modular Python functions.
- Geometric Alignment (K-Means & Agglomerative): Both algorithms achieved strong metrics due to the roughly spherical geometry of the feature space. Ward's linkage effectively minimized within-cluster variance.
- Density Realism (HDBSCAN): HDBSCAN refused to force overlapping structures into artificial partitions, identifying two dense natural continuums and dynamically isolating boundary anomalies.
Clustering Using K-Means, Agglomerative, and HDBSCAN.ipynb- The primary production-grade Jupyter Notebook containing the full benchmarking analysis.
- Language: Python
- Data Engineering: Pandas, NumPy
- Machine Learning & Modeling: Scikit-Learn (
KMeans,AgglomerativeClustering,StandardScaler), HDBSCAN - Data Visualization: Seaborn, Matplotlib
This project is licensed under the MIT License - see the LICENSE file for details.