Classification Methodologies

Challenges

 

When handling large volumes of data it is important that they can be classified, to provide information of value with the purpose of giving an appointed business goal.


Approach

 

Segmentation algorithms solve the problem of assigning an individual’s universe to study by a set of segments or clusters with similar characteristics.
There are different techniques depending on the problem: whether you know the set of segments in which it is necessary to classify the study sample, or if the set of segments is unknown and must be estimated.

Grupo AIA
has various methods and algorithms for automated segmentation:

  • Supervised Learning
  • Supervised Segmentation
  • Unsupervised Segmentation
  • Semi supervised Segmentation

 

 

Supervised Learning

EVA is a supervised learning tool that uses a dialogue generated by the system, where the expert must select only one of the two options presented by trapping the decision criteria used. As a result, it generates a linear discriminant with the relative weights of each relevant variable, able to mimic the behaviour of the expert.

Supervised Segmentation

Supervised segmentation algorithms require a known data set (labelled) to classify individual’s universe. Algorithms are called known target, which, based on a-priori knowledge of the classification segment, classify the data set.
Optimisation techniques are applied to the set of individuals known and it is extrapolated to the universe of individuals not known.

Unsupervised Segmentation


The unsupervised segmentation algorithms are applied when the set of segment classification is not known, and therefore must be estimated on the whole universe of individuals. Algorithms with an unknown target, scilicet, there are no a-priori knowledge about the number or type of segments in which individuals classify the universe.

AIA Group methods:

ADAN, machine learning tool for conceptual data clustering

Some other techniques utilized by Grupo AIA are:

  • k-means
  • k-medoid
  • Manhattan distance
  • Binary Divisive Partitioning

 

 

Semi supervised Segmentation

Semi supervised Segmentation algorithms are based on the premise that it is easier and less costly to obtain labelled information than untagged information. Such kinds of algorithms provide a mathematical method for labelling the entire universe of individuals based on a smaller set of labelled data.
Semi supervised Segmentation algorithms convert the work concerning a representative sample of the universe of individuals with real data in a study of the total universe of individuals with estimated data.

Some of the Grupo AIA algorithms used in this type of problem are:

  • Generative Models
  • S3VMs
  • Co-Training
  • MultiView Algorithms

 


Benefits

 

  • Allows analysis and manipulation of large volumes of data and extracts information that otherwise would go unnoticed.
  • Help in the decision-making process, bunching collectives of information in groups with similar trends.
  • Detects behaviours between individuals based on established business variables.

Success Cases

 

Grupo AIA has successfully implemented segmentation algorithms in several areas:

  • Credit Scoring
  • Interpretation of Electrocardiograms
  • Patterns banks detection
  • Market Behaviours
  • Classification of individuals by socio-demographic parameters