Instruction.txt: How to Run the Spectral Clustering Application

---

Introduction

This application demonstrates spectral clustering on different datasets using MATLAB 2024. The code includes interactive prompts for parameter selection and visualizes the clustering results.

---

Steps to Run the Application

1. Launch MATLAB 2024: Ensure you are running MATLAB version 2024 to guarantee compatibility with the code.

2. Load the Application: Save the provided code as a .m file (e.g., SpectralClusteringApp.m). Ensure all required dataset files (Circle.csv, Spiral.csv, Sphere.csv) are in the same directory as the script.

3. Run the Code: Open the .m file in MATLAB, then click on the Run button or type SpectralClusteringApp in the MATLAB command window.

4. Select a Dataset: The program will prompt you to choose one of the following datasets:

   • Circle.csv
   • Spiral.csv
   • Sphere.csv
     Use the menu dialog to select your desired dataset.

5. Set Parameters:

   • k (k-nearest neighbors): Enter the number of nearest neighbors (e.g., 10, 20, 40).
   • Laplacian Choice: Select the type of Laplacian matrix:
     • L: Standard Laplacian
     • L_norm: Normalized Laplacian
   • Number of Eigenvalues: Enter the number of eigenvalues to compute (e.g., 3, 5, etc.).

6. View Sparsity Pattern: The adjacency matrix's sparsity pattern is displayed in a plot for verification.

7. Eigenvalue Plot: After computing eigenvalues, the program plots them for analysis. Use this plot to identify the eigengap and determine the optimal number of clusters.

   • Tip: When prompted, enter the number of clusters based on the largest gap between consecutive eigenvalues.

8. Clustering Results: The script performs clustering using three methods:

   • k-means
   • DBSCAN
   • Hierarchical Clustering

9. Visualize Results: The program displays:

   • Clusters in 2D or 3D depending on the dataset.
   • Sparsity patterns of adjacency matrices.
   • Eigenvalue distributions.

10. Evaluate Silhouette Scores: The program computes and displays silhouette scores for each clustering method. These scores indicate how well-separated the clusters are.

11. Evaluate Adjusted Rand Index (ARI): If you select the Spiral.csv dataset, the ARI metric evaluates clustering accuracy against true labels.

---

Customizing the Clustering

• You can try different values of k, eigenvalues, and the number of clusters to see their effects on the results.
• Experiment with the DBSCAN epsilon parameter using the k-Distance graph generated by the code.

---

Note

• The datasets should be properly formatted with coordinates (and labels for Spiral.csv).
• Make sure MATLAB has the required permissions to access the datasets.