Classification of Data Mining Systems

Classification of Data Mining Systems

Data mining systems are classified along several orthogonal criteria, because a system is defined by what data it handles, what knowledge it finds, and how it finds it.

1. Classification Criteria (summary table)

Criterion	Classes	Examples
Kind of data mined	Relational, transactional, data-warehouse, object-relational / object-oriented, spatial, temporal & time-series, sequence, text, multimedia, web, stream, graph/social	spatial mining in GIS; clickstream mining
Kind of knowledge (functionality)	Characterisation & discrimination, association & correlation, classification, prediction/regression, clustering, outlier analysis, evolution analysis	Apriori, decision tree, k-means
Technique / technology used	Database-oriented / OLAP, machine learning, statistical, neural networks, genetic algorithms, fuzzy sets, rough sets, visualisation, agent based	ID3 (ML), regression (statistics), SOM (NN)
Level of abstraction	Primitive-level (raw) vs generalised/high-level knowledge; single-level vs multi-level (using concept hierarchies)	'young customers buy chips' is generalised
Data source & granularity	Source type (OLTP/WWW/sensor), data granularity (transaction/daily/summary), dimensionality, size	warehouse summary vs raw logs
Degree of user interaction	Query-driven (verification-driven), automatic / discovery-driven, interactive & exploratory, data-driven	OLAP drill vs automatic clustering
Application domain	Retail, finance & banking, telecom, insurance, healthcare & bioinformatics, manufacturing, security/intrusion detection, web/e-commerce	fraud detection, CRM churn

2. Data Mining vs Related Disciplines

                +----------------------------------+
                |   Database & Data Warehouse      |
                |   technology (SQL, OLAP, ETL)    |
                +----------------+-----------------+
                                 |
   +-----------------------------+-----------------------------+
   |   +----------------------+           +------------------+  |
   |   | Machine learning,    |  DATA     | Statistics,      |  |
   |   | pattern recognition, |  MINING   | probability,     |  |
   |   | neural nets, AI      |  (core)   | data analysis    |  |
   |   +----------------------+           +------------------+  |
   |        +------------------------+  +---------------------+ |
   |        | Information retrieval, |  | High performance /  | |
   |        | NLP, web technology    |  | distributed computing| |
   |        +------------------------+  +---------------------+ |
   +------------------------------------------------------------+

Related field	Contribution to data mining
Statistics	Sampling, hypothesis testing, regression, Bayesian inference, outlier tests
Machine learning / AI	Search strategies, induction algorithms, decision trees, SVM, neural nets, reinforcement learning
Database / data warehouse	Efficient storage, indexing, query optimisation, OLAP (multidimensional view), scalability
Information retrieval / NLP	Text indexing, similarity, web search, document clustering
Pattern recognition	Feature extraction, classification, clustering, image/audio mining
High-performance computing	Parallel, grid and GPU mining; distributed (Hadoop/Spark) algorithms
Visualisation / HCI	Visual data mining, interactive exploration, explanation of patterns

3. Descriptive vs Predictive Systems

Aspect	Descriptive mining	Predictive mining
Goal	Characterise existing data	Infer on unseen data
Functions	Clustering, association, characterisation, evolution	Classification, regression, prediction
Supervised?	No (unlabeled)	Usually yes (labelled target)
Example	'Customers who buy bread also buy butter'	'Will this customer default?'

4. Query-Driven vs Discovery-Driven

• Query-driven (verification-driven): the user asks a specific question and the system verifies/answers it — e.g. OLAP drill-down, SQL aggregation, 'is income related to churn?'.
• Data-driven (discovery-driven): the system searches automatically and returns surprising patterns — e.g. 'unexpected interaction between drug A and B'. Requires interestingness measures and pruning to avoid information overload.

5. Multi-Level / Multi-Dimensional Classification View

A mining system can be described as a point in a 4-dimensional space:

1. Data to be mined (which kind of source),
2. Knowledge to be discovered (which pattern type),
3. Technique used (which algorithm family),
4. Coupling / architecture (where mining sits relative to the DBMS).

6. Exam Points

• Explain the classification of data mining systems with examples — 10 marks (very frequent).
• Compare data mining with machine learning, statistics and OLAP — 5-7 marks.
• Distinguish descriptive vs predictive mining, and query-driven vs discovery-driven mining — 5 marks.
• Show inputs/outputs of mining visually with a diagram — 3-5 marks.