Performance Benchmarking of Clustering Methods using MNIST Data

This study aims to evaluate the effectiveness of three algorithms in clustering handwritten digit images from the MNIST dataset. The clustering algorithms assessed in this study are K-Means, Agglomerative Clustering, and GMM. The evaluation framework employs five assessment metrics: Silhouette Score, Calinski-Harabasz Score, Davies-Bouldin Score, Adjusted Rand Index, and Normalized Mutual Information Score. The computational complexity of each algorithm is analyzed by examining their runtime and resource consumption. The findings indicate that each algorithm has distinct strengths and weaknesses when applied to high-dimensional image data. K-Means demonstrated a balanced performance across all metrics, while Agglomerative Clustering excelled in producing well-defined clusters. GMM showed superior performance in capturing complex data distributions. The computational analysis revealed that K-Means is the most efficient in terms of runtime, whereas GMM requires more computational resources. Conclusion: The study concludes that while K-Means offers a good balance of performance and efficiency, Agglomerative Clustering and GMM provide better clustering quality for specific scenarios. These insights are valuable for selecting appropriate clustering methods for various data exploration and machine learning applications.