MSDN Documentation

Understanding Mining Model Content

This document provides a detailed explanation of how to interpret the content of mining models created in SQL Server Analysis Services (SSAS). Understanding the structure and meaning of model content is crucial for extracting meaningful insights and making accurate predictions.

Overview of Model Content

Each mining model in SSAS contains information specific to the algorithm used to create it. This content is stored in a hierarchical structure that can be queried using the DISCOVER_AMINING_MODEL_CONTENT rowset, or visualized through SQL Server Management Studio (SSMS) or SQL Server Data Tools (SSDT).

Common Elements of Mining Model Content

While specific content varies by algorithm, several common elements are found across most mining models:

• Node ID: A unique identifier for each node within the mining model structure.
• Parent Node ID: The identifier of the parent node in the hierarchy.
• Node Type: Indicates the type of node (e.g., root, branch, leaf, attribute, etc.).
• Node Caption: A human-readable description of the node.
• Properties: A collection of key-value pairs describing the node (e.g., probability, support, characteristic, value).
• Child IDs: Identifiers of child nodes associated with this node.

Algorithm-Specific Content

The richness of model content is best illustrated by examining specific algorithms:

1. Decision Tree Models

Decision tree models represent a series of rules that partition the data. The content includes:

• Splits: Nodes that represent a condition used to partition the data. The Properties field contains information about the attribute and its value range or specific values.
• Leaves: Terminal nodes that represent the predicted outcome or a characteristic associated with a specific path through the tree. The Properties field often includes the predicted value and its probability or confidence.

Example of a split node property:

{ "PropertyName": "CharacteristicName", "Value": "Age" }
{ "PropertyName": "Operator", "Value": "<=" }
{ "PropertyName": "Value", "Value": "35" }

2. Clustering Models

Clustering models group similar data points into clusters. The content describes:

• Cluster Name: A name or ID for the cluster.
• Cluster Characteristics: Attributes and their distribution within the cluster. The Properties field will contain information about attributes that are strongly associated with the cluster.
• Cluster Probabilities: The likelihood of a data point belonging to this cluster.

Example of cluster characteristics:

{ "PropertyName": "AttributeName", "Value": "Income" }
{ "PropertyName": "AttributeName", "Value": "Region" }
{ "PropertyName": "Support", "Value": 0.25 }

3. Association Rule Models

Association rule models discover relationships between items in transactional data. The content typically includes:

• Itemsets: Sets of items that frequently appear together.
• Rules: Implications derived from itemsets, showing relationships like "If {A} then {B}". The content will detail the support (frequency of the antecedent and consequent), confidence (probability of the consequent given the antecedent), and lift (how much more likely the consequent is given the antecedent than by chance).

Querying Mining Model Content

You can retrieve mining model content using DAX or AMO. The primary method is through Analysis Services' XML for Analysis (XMLA) Discover commands targeting specific rowsets, such as:

• DISCOVER_MINING_MODEL_CONTENT: Retrieves the content of a specific mining model.
• DISCOVER_ALL_MINING_MODELS: Lists all mining models in a database.
• DISCOVER_ALL_MINING_STRUCTURES: Lists all mining structures in a database.

A typical DMX query to retrieve content might look like this:

SELECT
    NODE_UNIQUE_NAME,
    NODE_CAPTION,
    NODE_TYPE,
    RELATIONSHIP_TYPE,
    CONTENT,
    CHILD_NODE_ID,
    PARENT_NODE_ID
FROM
    [YourModelName].CONTENT
WHERE
    NODE_TYPE = 20 -- Example: Leaf node type

Visualizing Model Content

SSMS and SSDT provide visual tools to explore mining model content. These tools translate the raw data into intuitive diagrams, trees, and tables, making it easier to understand the patterns and relationships discovered by the model.

• Decision Trees: Rendered as interactive trees.
• Clustering Models: Shown with cluster profiles and characteristic distributions.
• Association Rules: Presented as graphs or lists of rules with their metrics.

Conclusion

Mastering the interpretation of mining model content is a key step in leveraging the power of data mining with SQL Server Analysis Services. By understanding the structure and specific details provided for each algorithm, you can gain deeper insights into your data and build more effective predictive solutions.