Hors-Série IA 2020


Segmentation follows the same objectives as discriminant analysis but it is applied when explanatory variables are qualitative. The segmentation method consists on dividing the population on homogeneous sub-groups regarding the variable studied.
Segmentation process is iterative which means that at each stage the algorithm chooses the explanatory variable that correlates the best with the variable to explain in order to divide the population.
The division identifies the two segments with the highest inter-segment variance and the lowest intra-segment variance. The result of segmentation process is a sort of decision tree with a division of each group into two sub-groups. The first division enables the researcher to obtain the two first sub-groups. Each one of these sub-groups are then divided in two sub-groups with the variable allowing the best division (generally it is not the same for both groups). The process continues with some interruptions in the case in which the size of a group falls below a specified threshold, or when the low variance percentage is due to the optimum division.