🤖 AI / ML

Unsupervised Learning

unsupervised learning · unlabelled data · unlabeled learning · clustering · k-means · dimensionality reduction · anomaly detection

Definition

Unsupervised learning is a type of machine learning in which a model discovers structures, clusters or representations in data without using labelled target values.

What unsupervised learning is for

In most e-commerce problems the data is not labelled: there are millions of events — views,
add-to-basket actions, purchases — but no explicit answer to “which segment is this shopper in?” or
“is this transaction anomalous?”. Unsupervised learning finds the hidden structure in such data by
itself.

The three main tasks

Clustering — splitting objects into groups with high internal similarity. Shoppers with similar
behaviour land in the same cluster. The result is a set of audience segments that can each be given
a different strategy.

Dimensionality reduction — compressing high-dimensional data, such as a purchase vector across
50,000 products, into a compact representation of 10 to 50 features. The algorithms: PCA, t-SNE,
UMAP, autoencoders. Used for visualising segments and for preprocessing before training.

Anomaly detection — finding the points that do not fit the overall structure. In e-commerce:
fraudulent orders, review manipulation, bot traffic.

Use in audience segmentation

Clustering on RFM (recency, frequency, monetary) is the classic example. K-means — or DBSCAN when
the clusters have an irregular shape — divides shoppers into segments with no rules set in advance:

Cluster 1: high F, high M          → VIP customers
Cluster 2: recent R, low F         → new customers
Cluster 3: distant R, formerly high F → churn risk

The output is a set of segments usable for content personalization, triggered communications and
differentiated offers.

Tip: clustering results have to be interpreted by hand. The algorithm will find the groups —
but naming them (“loyal”, “churning”, “new”) is the analyst’s job. Without a business reading, the
clusters stay just numbers.

Frequently asked questions

How is unsupervised learning different from supervised learning?

In supervised learning every example carries a correct answer — a label: the product was bought or it was not, the customer churned or stayed. The algorithm learns to predict that answer. In unsupervised learning there are no labels, so the algorithm searches for hidden structure on its own, working from the raw data alone.

Which e-commerce tasks does unsupervised learning solve?

Clustering shoppers by behaviour and RFM to build audience segments. Building thematic product clusters to improve navigation and recommendations. Anomaly detection for fraudulent transactions or bot traffic. Dimensionality reduction of embeddings for visualisation and faster retrieval.

What is k-means and when should you use it?

K-means is the simplest clustering algorithm. It splits the data into K clusters by minimising the distance from every point to the centroid of its cluster. It suits customer segmentation on numeric features — RFM, order value, frequency. Its limitation is the assumption that clusters are spherical and of similar size, which real audiences often are not.

Unsupervised Learning

What unsupervised learning is for

The three main tasks

Use in audience segmentation

Frequently asked questions

Related terms

We received your request