Segmentation

We are drowning in an ocean of information, and it is very common not knowing exactly what we are looking for in data. Setting well-defined categories or establishing fixed measurements according to which classify and organise data may be a very complex task, despite having in mind a clear objective. “I want to know my customers better” is a very frequent objective, but questions related to it, such as “What interest groups are there among my customers?”, still have no simple answer.

Segmentation techniques analyse high volumes of data and detect automatically related groups of elements, or even opposing cases that do not follow the standards. They are tools useful to:

Automatic Segmentation is grounded on the scientific discipline known as unsupervised machine learning. This transforms data into a numerical representation that allows calculating similarities between elements and detecting related groups or elements that are far more different than expected.

A process of segmentation results in a list of related groups detected in the data to be later studied by the analysts of the client. Segmentation may be carried out many times, whenever data under study are updated, which allows finding out changes on the groups identified, such as the emergence of new interests in a database of customers, or new topics in a set of documents.

As far as the service of detection of anomalies is concerned, the result of the segmentation process is a model of detection which can be displayed to identify anomalous data or behaviours in real time. As is the case with classification systems, the detection model generates a certainty score that indicates the degree of irregularity identified, so it would be possible to set business rules, for instance, to control accesses to a network and block automatically the ones showing a high score.