Identifying data segments

Umnai can identify data segments by inspecting the rules created during induction.

During induction, the training dataset is segmented into regions that behave similarly. These segments are encoded in the conditional_expression of each rule.

This information helps explain both the dataset and the model: it shows how Umnai grouped data by behaviour, and how those groups relate to the segments people already use in the domain.

How segments are represented

Rules describe the conditions that activate a partition.

The conditional_expression of a rule defines the segment of data associated with that rule. For example, an Age rule may define a range such as:

0 < Age <= 23

When a query falls within that range, the corresponding rule and partition can activate.

Looking across the rules in a module shows how induction segmented that feature or interaction into regions of similar behaviour.

Comparing discovered segments with business segments

Discovered segments can be compared with existing business, operational, or domain-defined segments.

For example, a marketing team may use standard Age groups:

The Age module rules may identify different behavioural bands:

The difference between these segmentations can be useful. It may show that the model has learned behavioural boundaries that do not match the organisation’s existing segmentation.

Using segment insights

Segment insights can support business and model review.

If the discovered segments make sense, they may strengthen confidence in the model and in the dataset. If they differ from expected business segments, they may prompt further investigation.

For example, the marketing team may re-evaluate its market segmentation and campaigns if the model consistently identifies 0 to 23, 23 to 55, and over 55 as behaviourally meaningful groups rather than the existing 0 to 18, 18 to 65, and over 65 groups.

This does not automatically mean the original business segmentation is wrong. It provides evidence that can be compared with domain knowledge, business strategy, and additional data.