What is Unsupervised Learning?
Unsupervised learning is a machine learning approach that finds structure in data carrying no labels. Instead of predicting a known answer, algorithms group similar items, reduce dimensionality, estimate density, or detect items that do not fit the observed pattern. Results describe the data rather than being correct or incorrect against a supplied target.
Common families include clustering, which partitions items into groups by similarity, dimensionality reduction, which compresses many variables into a few informative ones, and anomaly detection, which flags points far from the bulk of a distribution. Association rule mining, which surfaces items that frequently occur together, belongs to the same category. Each answers a descriptive question rather than a predictive one.
Because there is no ground truth, evaluation is indirect. Analysts use internal measures of cluster cohesion and separation, stability under resampling, and usefulness for a downstream task, and they inspect results manually. The number of clusters, the distance measure, and the scaling of features all change the answer, so unsupervised results are best treated as hypotheses requiring validation.
Self supervised learning is closely related and is how modern language models are pretrained. The training signal comes from the data itself, for example by predicting a hidden or next token, so no human annotation is needed. Some authors treat this as a subtype of unsupervised learning, while others separate it because an explicit prediction target is constructed.
Practical uses include customer segmentation, topic discovery in document collections, compression of high dimensional measurements before supervised modeling, and monitoring systems that flag unusual behavior. Embeddings, which place items in a vector space where proximity reflects similarity, are frequently learned without labels and then reused by search, recommendation, and retrieval systems.
Key points
- Finds structure in unlabeled data.
- Includes clustering, dimensionality reduction, and anomaly detection.
- No ground truth, so evaluation is indirect.
- Self supervised pretraining is a closely related method.
- Results are hypotheses that still need validation.
In practice
An analyst has purchase histories with no customer segments defined. Clustering groups shoppers whose baskets and timing resemble one another, and several coherent groups appear, such as frequent small orders and rare bulk orders. Nothing labeled these groups in advance. The analyst inspects them, names them, and only then treats the segmentation as a business input.