Cluster ranking
When Nuance Gatekeeper returns the clustering results, it uses a set of data to rank them such that the most likely fraudulent cases appear at the top of the list. This ranking is defined by the combination of criteria weight and the ranking of the result for each criterion.
The following table describes the cluster ranking criteria:
| Priority | Weight | Criterion | Value type | Description |
|---|---|---|---|---|
| 1 | 5 | Person ID = different | Boolean | Whether the voice in the sample corresponds to the expected voiceprint |
| 2 | 4 | Verdict = fraud | Boolean | Whether the system has reached a verdict that the current cluster is fraudulent |
| 3 | 3 | Average risk | Float | The risk factor associated to the cluster |
| 4 | 2 | Decision = fraud | Boolean | Whether there is a decision that the current cluster is fraudulent |
| 5 | 1 | ANI = different | Boolean | If the phone number for the current cluster is different from the identified user’s registered phone number |
| 6 | 1 | ANI = unknown/blocked | Boolean | If the phone number for the current cluster is unavailable due to it being unknown or blocked |
| 7 | 1 | Purity | float | The cluster’s purity score |
The ranking weight for each of the criteria is configurable by the user.
The cluster ranking is calculated by multiplying the value of the criterion by the ranking weight, and then adding the values for the cluster. The result of the calculation is a floating point number that can then be used to determine how to rank each individual cluster against the other.
Ranking example
In this example there are two clusters being ranked.
Cluster 1
The following table lists the ranking criteria of Cluster 1:
| Priority | Weight | Criterion | Value |
|---|---|---|---|
| 1 | 5 | Person ID = different | 1 |
| 2 | 4 | Verdict = fraud | 0 |
| 3 | 3 | Average risk | 149/1000 = 0.149 |
| 4 | 2 | Decision = fraud | 0 |
| 5 | 1 | ANI = different | 0 |
| 6 | 1 | ANI = unknown/blocked | 0 |
| 7 | 1 | Purity | 6.6/10 = 0.66 |
The ranking for the cluster is the result of this formula: (5x1) + (4x0) + (3x0.149) + (2x0) + (1x0) + (1x0) + (1x0.66) = 6.11
Cluster 2
The following table lists the ranking criteria of Cluster 2:
| Priority | Weight | Criterion | Value |
|---|---|---|---|
| 1 | 5 | Person ID = different | 0 |
| 2 | 4 | Verdict = fraud | 1 |
| 3 | 3 | Average risk | 448/1000 = 0.448 |
| 4 | 2 | Decision = fraud | 1 |
| 5 | 1 | ANI = different | 0 |
| 6 | 1 | ANI = unknown/blocked | 0 |
| 7 | 1 | Purity | 7.5/10 = 0.75 |
The ranking for the cluster is the result of this formula: (5x0) + (4x1) + (3x0.448) + (2x1) + (1x0) + (1x0) + (1x0.75) = 8.09
Result
In this example, Cluster 2 (8.09) will be ranked higher than Cluster 1 (6.11).