Auto-Clustering

Framework

Problem

Answering the frequent Client question:

Know the behaviour to prevent deviations from the standard behaviour of its Products/Services/Customers”.

Humanativa – Tools and Expertise

Clustering

Clustering is a technique of subdividing a population or ‘data points’ by grouping them into different clusters on the basis of similarity and dissimilarity between them.

Clustering processes are widely known to the specialist community, and a fair variety of algorithms exist in the literature, but in Humanativa we apply a two-level clustering criterion, refining clusters for ‘robust’ and self-consistent behavioural patterns.

We fill a gap in the market by offering a rigorous protocol that, starting from literature algorithms, performs a grid-search, analogous to the one most widely used in the field of supervised problems, in order to compete several algorithms in several parameter configurations and identify which is the most ‘efficient’ among the competing models.

Benefits

  • Understanding Customer/Product Segmentations and how the ‘population’ of segments moves over time through multiple observations of patterns.
  • Very useful in the Data Preparation (ML) phase to understand the Value of Data right from the Analysis phase.

Use Case

  • Industry: Banking
    Use case: Securitisation Analysis (NPL) – Customer insight for Up – Cross selling – Leasing Analytics – Credit Risk
  • Industry: Automotive – Smartcity
    Use case: Behavioural profiling in Vehicle Flow – Behavioural analytics & accident risks
  • Industry: Energy
    Use Case: Credit Collection