Data Mining Functionalities

1. What is Data Mining?

Data mining is the non-trivial process of extracting implicit, previously unknown and potentially useful patterns and knowledge from large volumes of data.

Alternative names: KDD (Knowledge Discovery in Databases), knowledge extraction, data/pattern analysis, data dredging, information harvesting, business intelligence.

2. KDD Process vs Data Mining

 +----------+   +----------+   +------------------+   +-------------+
 | RAW DATA |-> | SELECTION|-> | PREPROCESSING    |-> | TRANSFORM-  |
 +----------+   | (target  |   | (cleaning,       |   | ATION       |
                |  data)   |   |  integration)    |   | (features,  |
                +----------+   +------------------+   |  discreti-  |
                                                      |  zation)    |
                                                      +------+------+ 
                                                             |
        +-------------------+    +----------------+   +-----v------+
        | KNOWLEDGE         |<-- | PATTERN        |<--| DATA       |
        | (patterns, rules, |    | EVALUATION     |   | MINING     |
        |  visualisation)   |    | (interesting-  |   | (algorithms|
        +-------------------+    |  ness, support,|   |  + search) |
                                 |  confidence)   |   +------------+
                                 +----------------+

Data mining is one step (the core step) of the KDD process, not the whole process.

3. Major Data Mining Functionalities

#	Functionality	Goal	Typical techniques
1	Concept / class description	Summarise data — characterization (target class) & discrimination (contrast classes)	Attribute-oriented induction, OLAP roll-up, decision-rule characterisation
2	Frequent pattern, association & correlation mining	Find items that co-occur	Apriori, FP-growth, Eclat, correlation rules
3	Classification	Predict a discrete class label	Decision tree, naive Bayes, k-NN, SVM, neural net, rule induction
4	Prediction / regression	Predict a continuous value	Linear/logistic regression, ARIMA, neural nets
5	Cluster analysis	Group similar objects (unsupervised)	k-means, k-medoids, hierarchical, DBSCAN, SOM
6	Outlier / anomaly analysis	Detect rare, unusual objects	Statistical tests, distance-based, density-based (LOF)
7	Evolution analysis	Model trends over time	Time-series analysis, sequence mining, trend/periodicity detection

Descriptive vs predictive view:

 DESCRIPTIVE (characterise existing data)        PREDICTIVE (infer on new data)
 - concept description                            - classification
 - association / correlation                      - regression / prediction
 - clustering                                     - outlier prediction
 - evolution / trend analysis                     - sequence / trend prediction

4. Types of Data that can be Mined

Relational DBs, data warehouses, transactional DBs, object-relational & object-oriented DBs, spatial & temporal data, time-series & sequence data, text, multimedia (image/audio/video), World Wide Web, data streams, graph & social-network data.

5. Interestingness of Patterns

A pattern is interesting if it is:

Non-trivial, implicit, previously unknown or unexpected, and potentially useful.
Objective measures: support, confidence, lift, correlation, chi-square, Gini.
Subjective measures: surprise, novelty, actionability — depend on the user's belief and business goal.

\text{support}(A \Rightarrow B) = P(A \cup B), \qquad \text{confidence}(A \Rightarrow B) = \frac{P(A \cup B)}{P(A)}, \qquad \text{lift} = \frac{P(A \cup B)}{P(A)P(B)}

A rule is strong if support ≥ min_sup and confidence ≥ min_conf; lift > 1 means positive correlation, lift = 1 independence, lift < 1 negative correlation.

6. Requirements of a Data Mining System

Handle large volumes (GB-TB) and high dimensionality.
Efficient, scalable algorithms; user interaction and prior knowledge.
Interestingness measures and pattern pruning; robustness to noise and missing values.
Integration with DBMS/warehouse, security, privacy protection.

7. Applications

Market basket analysis, customer segmentation & churn, credit scoring & fraud detection, medical diagnosis, bioinformatics, telecom network fault analysis, intrusion detection, web log / clickstream mining, recommendation engines, manufacturing quality control.

8. Exam Points

Draw and explain the KDD process and state where data mining fits — 10 marks (very frequent).
Explain any five data mining functionalities with examples — 10 marks.
Define interestingness, support, confidence, lift with formula — 5 marks.
Distinguish KDD and data mining — 3 marks.

Data Mining Functionalities