|
|
Link Analysis Node |
You can use the link analysis node for visual data mining by examining relationships between values within different fields. This feature is frequently used for market-basket analysis.
The two fields that are being analysed are first either discretised into a specified number of bins for numerical values, or divided into bins of unique values for categorical fields. A pair-wise comparison of all possible combinations of these bins is then performed and the number of records falling within each pair of bins is determined.
Results can be displayed in either a graphical format (sometimes referred to a web graph) or in tabular form in an HTML report. The tabular form is particularly beneficial where a large number of fields are available, as all of the relationships can be viewed in a single report (making it a rapid method of identifying links between field values).
A relationship between two field values is indicated on the graph by a line connecting two points that represent the values. The strength of the relationship is shown by the colour and thickness of the joining line. You can configure the settings for three line types based on the number of records that support each relationship.
For example: If the first field is numerical (ranging from 0 to 32) and the second is categorical (with the 2 values "YES" and "NO") then, with discretisation set to the default value of 10 bins, the following pairs of bins would be analysed. The values in each cell indicate the number of records in the database that fall within both bins.
|
| ||||||||||
|
Field1 →
↓ Field2 |
0.00 - 3.20 |
3.20 - 6.40 |
6.40 - 9.60 |
9.60 - 12.80 |
12.80 - 16.00 |
16.00 - 19.20 |
19.20 - 22.40 |
22.40 - 25.60 |
25.60 - 28.80 |
28.80 - 32.00 |
|
NO |
454 |
6 |
140 |
56 |
94 |
141 |
0 |
0 |
0 |
1 |
|
YES |
1 |
0 |
34 |
18 |
27 |
37 |
2 |
1 |
0 |
0 |
This analysis can also be displayed graphically. In such cases the strength of the relationships that exist are indicated by lines of a user-definable width in the approriate color.
Such anaysis can be used to provide a simple and efficient method of visualizing the structure of the entire database, often enabling interesting patterns within the data to be rapidly determined.
Full details of the options available for the link analysis node can be found on the link analysis options page.
