For decades, speaker diarization techniques have explored various methodologies, with clustering algorithms playing a crucial role in distinguishing speakers. While traditional methods such as PCA and SVD are commonly used, their effectiveness in complex multi-speaker scenarios warrants further investigation. In this study, we evaluated eight different dimensionality reduction and clustering methods for the speaker diarization task using MFCC features. Specifically, we compare PCA, Truncated SVD, Incremental PCA, Kernel PCA with RBF kernel, t-SNE, Locally Linear Embedding, Isomap, and UMAP. MFCC features were extracted and clustering methods were applied. The results indicate that UMAP significantly outperforms the other techniques, with an ARI of 0.89 and a Silhouette Score of 0.74, while other methods, such as PCA and Truncated SVD, achieve an ARI of only 0.27. Notably, kernel PCA with RBF kernel proved entirely ineffective, while t-SNE and Locally Linear Embedding yielded ARI values close to zero. This research demonstrates the potential of UMAP in speaker diarization applications and opens up avenues for integrating UMAP with deep learning techniques to further enhance clustering performance.