Where data mining is applicable

Areas where data mining is applicable

Data mining is useful where there is a large amount of data and a need to find patterns:

• Retail and marketing: market-basket analysis, customer segmentation, campaign response.
• Banking and finance: credit scoring, fraud detection, risk analysis.
• Telecom: churn prediction, usage analysis.
• Healthcare and bioinformatics: diagnosis, gene analysis, drug discovery.
• Web and e-commerce: recommendation, web usage mining, search.
• Science and engineering: weather, astronomy, fault prediction.
• Government and security: crime analysis, tax-evasion detection.

Reason: the data is large, patterns are hidden, and the discovered knowledge gives business value.

Areas where it is not applicable

• Very small data sets: patterns are not statistically reliable; simple statistics suffice.
• When the answer is known by simple queries or rules: SQL or reports give the answer directly.
• Poor-quality or irrelevant data: noisy, biased or incomplete data gives misleading results ("garbage in, garbage out").
• No clear business goal or no way to act on the findings.
• Cases needing exact guarantees (e.g. safety-critical decisions that cannot tolerate probabilistic results).
• Privacy or legal restrictions that prevent using the data.

Is a data warehouse compulsory for data mining?

No. Data mining can run on flat files, a relational database, or any data repository. But a warehouse is highly desirable: it provides clean, integrated, consolidated and historical data, which saves 60 to 80% of the preparation effort and improves mining quality. Without it, the mining team must do the cleaning and integration itself.