Categories of Functions in Data Mining

Nearly every specific technique covered throughout this series — clustering, association rules, classification, correlation analysis — fits into an organizing taxonomy, and understanding that taxonomy is more useful than re-deriving each technique a second time. This lesson is that map: what the two broad categories of data mining functionality are, which specific functions fall under each, and — a distinction the source material for this topic frequently blurs — why "categories of functions" and "task primitives" are two entirely different ideas that happen to get listed side by side.

The Two Broad Categories

Every data mining functionality answers one of two fundamentally different kinds of question, covered with a full comparison in the lesson on descriptive and predictive data mining:

• Descriptive functions characterize properties of the data that already exists — what patterns, groupings, or relationships are present right now.
• Predictive functions estimate a value that isn't yet known — a future outcome, or a label for a new, unseen case — based on patterns learned from existing data.

What Falls Under Each Category

Under descriptive mining:

• Characterization and discrimination — summarizing a target class's typical properties, or comparing two classes against each other, covered with worked examples in the lesson on class comparison.
• Clustering — grouping similar records together without predefined labels, covered throughout the clustering-focused lessons in this series, including the grid-based approach covered in its own dedicated lesson.
• Frequent patterns, association, and correlation analysis — finding itemsets, sequences, or variables that co-occur or move together more than chance would predict, covered in full depth (support, confidence, lift, leverage) in the lesson on association rule mining and the lesson on correlation analysis. Under predictive mining:
• Classification — assigning a new case to one of several predefined categories, covered with a fully worked example in the lesson on classification by backpropagation.
• Regression and numeric prediction — estimating a continuous numeric value rather than a category, covered in the lesson on forecasting.

Task Primitives Are a Different Thing Entirely

It's worth correcting a genuine point of confusion here: task primitives are not another category of function sitting alongside characterization, clustering, and classification — they're the specification framework used to configure any of those functions for a specific request. The five primitives (task-relevant data, the kind of knowledge to mine, background knowledge, interestingness measures, and presentation) answer "which data, which function, which threshold, shown how" — not "which function." This distinction is exactly what DMQL's clause structure makes concrete, covered in the lesson on the data mining query language: the mine <kind_of_knowledge> clause is where a task primitive selects one of the functions from this lesson's taxonomy (mine associations, mine classification), while with <measure> threshold = <value> is a separate primitive specifying the interestingness cutoff — two different primitives configuring one chosen function, not two functions in the same list.

Matching a Question to a Function

The practical value of this taxonomy is being able to go from a business question to the right function quickly:

Every row in this table is a different function answering a structurally different question — and the reason so many of them are covered in dedicated lessons elsewhere in this series, rather than derived again here, is that this lesson's job is the map, not the territory.