Unsupervised learning (UL) serves as the cornerstone for exploring data and discovering hidden patterns without labeled examples, and this paper delivers a rigorous, technically deep survey covering foundational concepts like density estimation and latent variable models. The survey comprehensively reviews clustering algorithms such as K-means and DBSCAN, dimensionality reduction techniques like PCA, UMAP, and autoencoders, alongside association rule mining and anomaly detection methods like isolation forests. Furthermore, it analyzes key evaluation metrics including the silhouette score and reconstruction loss, while highlighting modern advances such as contrastive learning, VAEs, and GANs. Ultimately, the text presents real-world case studies spanning customer segmentation and recommender systems, providing a practical decision framework for algorithm selection based on data characteristics along with a benchmarking methodology for fair comparison.