Note

This repository currently contains core files. A full release with complete code, configs, and documentation will follow with the paper’s preprint/publication.

Streaming k-Medoids for Mixed-Type Healthcare Data

This repository contains the code, configuration files, and scripts used in the paper:

Streaming + Coreset k-Medoids for Large-Scale Mixed-Type Healthcare Data
(Submitted to NeurIPS 2025 MusIML Workshop)

---

🔧 Features

• Streaming + coreset k-medoids algorithm with weighted Gower distance
• Supports numeric, binary, and categorical features
• Chunk-wise streaming with Hungarian alignment of medoids
• Coreset refinement for scalability under strict memory limits
• Feature weighting modes: uniform, manual (e.g., ethnicity emphasis), supervised (benchmark only)
• Built-in metrics: ARI, NMI, Silhouette, Purity, and cluster-specific Precision/Recall
• Logging of runtime, peak memory, and learned feature weights

💻 Environment

Python 3.10

📄 Citation

If you use this code, please cite:

@inproceedings{Shah2025StreamingKMedoids, title = {Streaming + Coreset k-Medoids for Large-Scale Mixed-Type Healthcare Data}, author = {Syed Ahmar Shah, Fatima Almaghrabi, Aziz Sheikh}, booktitle = {NeurIPS 2025 Muslims in ML Workshop}, year = {2025} }