Unsupervised

Unsupervised learning is a branch of machine learning in which models are trained on data without labeled responses. The goal is to uncover hidden structure, patterns, or representations in the data rather than to predict a defined target.

Common tasks include clustering to group similar observations, dimensionality reduction to simplify high-dimensional data, density estimation to model data distributions, and representation learning to capture meaningful features. Algorithms range from distance-based methods such as k-means and DBSCAN to dimensionality reduction like PCA, t-SNE, and UMAP; probabilistic models such as Gaussian mixture models; and neural approaches such as autoencoders and variational autoencoders.

Applications include market segmentation, anomaly detection, data compression, visualization, and preprocessing for supervised tasks where labeled data are scarce. In natural language processing and computer vision, unsupervised techniques are used to learn word or image representations from raw data.

Evaluation is challenging due to the absence of ground-truth labels and often relies on internal criteria (for clustering: silhouette score, Davies-Bouldin index) or on downstream performance after training a supervised model. Limitations include sensitivity to hyperparameters, potential to learn irrelevant structure, and difficulty ensuring interpretability.

Historically, early methods such as principal component analysis and k-means demonstrated the utility of unsupervised learning, which has since expanded into probabilistic models, deep learning-based representation learning, and self-supervised variants that use data-derived tasks to shape representations.