Interactive Clustering with GenAI

An interactive Streamlit-based application for performing traditional and deep learning-based clustering, enhanced with explainable AI (XAI) and Generative AI (GenAI)-powered insights.

---

Features

Clustering Techniques

• Centroid-based: KMeans
• Density-based: DBSCAN, HDBSCAN
• Connectivity-based: Agglomerative Hierarchical Clustering (Ward, Complete, Single, Average)
• Distribution-based: Gaussian Mixture Models (GMM)
• Grid-based: (Optional extension)
• Self-Organizing Maps (SOM)
• Deep Learning-based: Autoencoder + KMeans

Evaluation Metrics

• Silhouette Score
• Davies-Bouldin Index
• Calinski-Harabasz Index
• Adjusted Rand Index (ARI)
• Normalized Mutual Information (NMI)

Dimensionality Reduction & Visualization

• PCA, t-SNE, UMAP
• Visualize cluster separation, outliers, and centroids
• Supports both Matplotlib and Plotly for interactive visuals

GenAI Features

• Cluster Summary Generator: Uses Hugging Face API (google/flan-t5-base) to summarize cluster characteristics
• Accepts Hugging Face API key via input prompt
• Automatically generates human-readable insights based on cluster stats

---

How to Use

1. Clone this repository

git clone  origin https://github.com/sivkri/GenAI-cluster-analyzer           
cd GenAI-cluster-analyzer

2. Install dependencies

pip install -r requirements.txt

3. Run the Streamlit app

streamlit run streamlit_bot.py

App Flow

1. Upload your dataset (CSV format)

2. Preview the first 5 rows

3. Select columns to drop

4. Select features for clustering

5. Choose clustering method(s)

6. View evaluation metrics and interactive visualizations

7. Optionally generate a GenAI-powered cluster summary

---

Hugging Face API

To enable GenAI cluster summarization:

1. Go to Hugging Face Tokens

2. Generate a read access token

3. Paste your API key into the app when prompted