This repository showcases MATLAB-based projects focusing on custom implementations of clustering and classification algorithms for various datasets. Each project combines rigorous mathematical methodologies with comprehensive MATLAB coding to address real-world classification challenges, covering techniques such as K-means, K-medoids, and clustering evaluations. You can find the full description on the pdf file.
- Iris Dataset Clustering with K-means and K-medoids
- Objective: Classify the well-known Iris dataset into three species using custom implementations of K-means and K-medoids algorithms.
- Methodology: The K-means and K-medoids algorithms were implemented from scratch. K-means minimizes within-cluster variance, while K-medoids (using the L1 distance) reduces sensitivity to outliers by choosing actual data points as medoids.
- Evaluation: Results are evaluated using confusion matrices and misclassification counts across multiple runs, highlighting each algorithm’s stability and accuracy.
- Breast Cancer Biopsy Data Analysis with K-medoids
- Objective: Cluster biopsy data to distinguish between benign and malignant samples, focusing on maximizing accuracy and robustness.
- Methodology: Missing data entries are handled, and K-medoids clustering is employed, using sensitivity and specificity metrics to assess the method’s effectiveness in identifying malignant cases.
- Evaluation: Sensitivity and specificity scores reveal the clustering accuracy, making it a practical diagnostic tool. The algorithm is tested across multiple runs to ensure robust results.
- 1984 Congressional Voting Records Analysis
- Objective: Investigate partisan voting behavior in the 1984 U.S. Congress by clustering representatives based on their voting patterns.
- Methodology: A dissimilarity matrix was computed using a custom dissimilarity index for "yes"/"no" votes, while managing missing votes by assigning a neutral score. K-medoids clustering is then applied to group representatives by voting alignment.
- Evaluation: Confusion matrices reveal a clear partisan split, showing the effectiveness of K-medoids in political data clustering. Additional analysis assesses the voting consistency within each cluster.
- Wine Classification Using Chemical Properties
- Objective: Classify Italian wines from three cultivars based on 13 chemical attributes, comparing the performance of K-means and K-medoids algorithms.
- Methodology: Both K-means and K-medoids algorithms are applied to the wine dataset, with clusters evaluated for accuracy against the true wine cultivars.
- Evaluation: Confusion matrices are generated to identify clustering accuracy. This project highlights the clustering challenges posed by similar chemical profiles across cultivars and compares the resilience of each algorithm.
- Cardiac SPECT Data Clustering for Patient Classification
- Objective: Classify cardiac patients as normal or abnormal using binary data from SPECT images.
- Methodology: Dissimilarity matrices are created using binary metrics, followed by K-medoids clustering to separate patients based on diagnostic markers.
- Evaluation: Confusion matrices and a custom 2x2 classification matrix evaluate the clustering accuracy, reflecting how well attributes correspond to patient health status.
- Language: MATLAB
- Techniques: K-means clustering, K-medoids clustering, Confusion matrix analysis, Sensitivity and specificity scoring, Robustness testing
- Datasets: Iris, Biopsy, Congressional Voting, Wine, and Cardiac SPECT datasets
- Custom Implementations: All algorithms are coded from scratch to deepen understanding of clustering mechanics.
- Robust Evaluations: Each project includes evaluation metrics like confusion matrices and error analyses to measure clustering quality.
- Data Insights: These projects emphasize understanding algorithm strengths and limitations, particularly in handling outliers and noisy data.
- Each project is accompanied by detailed plots and figures, including:
- Cluster assignments with centroid/medoid visualizations
- Confusion matrices for classification accuracy
- Performance graphs illustrating algorithm convergence
To explore these projects:
- Clone this repository.
- Open MATLAB and navigate to the project folder.
- Run each script to reproduce the results, visualizations, and metrics outlined above.
git clone https://github.com/yourusername/MATLAB-Clustering-Classification.git
cd MATLAB-Clustering-Classification