UIC-Miner Task

The UIC-Miner (Uplift in Categories Miner) finds conditions under which rare categories in a target attribute are proportionally boosted. It is designed for situations where a valuable but infrequent outcome — such as a fatal accident, a high-value sale, or a rare disease — needs to be identified and amplified despite its low base occurrence.

The core idea: rather than searching for conditions where a rare category appears alone (which may produce too few results to be actionable), UIC-Miner mixes the rare category with a related, more common one and searches for conditions where the combined weighted improvement is above a defined threshold.

Task creation is handled through a three-step wizard: Task Setup → Logic Configuration → Quantifiers.

---

Step 1 — Task Setup

The first step captures the basic task information:

Field	Description
Task Name	A name to identify this task
Procedure Method	Select UICMiner
Dataset	The dataset to mine — select from your uploaded datasets
Project	Optionally assign the task to a project (can be left empty)

Click Next Step to proceed to logic configuration.

---

Step 2 — Logic Configuration

The UIC-Miner has two cedent tabs — Antecedent (IF) and Condition (Filter) — plus a Target Attribute selector shared with the CF-Miner.

Target Attribute

Select the categorical column whose category distribution you want to analyse and boost. The categories of this attribute are what the AAD Weights quantifier in Step 3 will be applied to.

Antecedent (IF)

The antecedent defines the conditions the miner explores. For each antecedent found, the miner evaluates the histogram of the target attribute within that subgroup and compares it to the baseline (either the full dataset or the condition subset).

Configure it the same way as any other cedent:

• Cedent Type — toggle between Conjunction (AND) or Disjunction (OR) using the Switch Type button
• Cedent Length (Min / Max) — controls how many attributes can be combined in a single antecedent

For each attribute added:

Field	Description
Column	Select an attribute from the dataset
Type	How the attribute's values are grouped — see Literal Types below
Min / Max	The minimum and maximum number of values to combine for this attribute

Condition (Filter)

An optional filter that restricts the dataset to a subset before mining begins. When set, the antecedent's histogram is compared against the condition subset rather than the full dataset.

Literal Types

Type	Description
subset	Any subset of the attribute's categories (unordered)
seq	Sequences of consecutive ordered values
lcut	Left cut — takes values from the left end of the ordered range
rcut	Right cut — takes values from the right end of the ordered range

Click Next Step to proceed to quantifier setup.

---

Step 3 — Quantifiers

UIC-Miner uses a weighted scoring mechanism to rank how interesting a discovered antecedent is. Only rules meeting all specified conditions are returned.

Base

Quantifier	Description
Base	Minimum number of records satisfying the antecedent
Relative Base	Minimum base as a fraction of the total dataset size

Confidence (AAD Score)

Quantifier	Description
AAD Score	Minimum weighted improvement score. Measures the overall uplift across all target categories, weighted by importance. Higher means more interesting rules.
AAD Weights	A comma-separated weight vector — one value per target category, in order. Categories with higher weights contribute more to the score. Set to 0 to exclude a category from scoring.

How AAD Weights work

The weights tell the miner which categories matter most. For example, with a target of Severity having categories Fatal, Serious, Slight in that order:

• 5, 1, 0 — Fatal is 5x more important than Serious, Slight is ignored
• 1, 1, 0 — Fatal and Serious matter equally, Slight is ignored
• 5, 0, 0 — only boost in Fatal counts

The miner computes the per-category improvement (how much the antecedent raises the occurrence of each category relative to the baseline), multiplies each by its weight, and sums them into a single aad_score. Only rules exceeding the AAD Score threshold are included in the results.

Submitting the Task

At the bottom of the quantifiers step, two actions are available:

• Save Task — saves the task configuration for later execution
• Run Task — saves the task and immediately dispatches it to the execution pipeline

tip

Start with a moderate AAD Score threshold (e.g. 20) and assign the highest weight to your rarest but most valuable category. You can always raise the threshold to tighten the result set after seeing the initial output.