PATTERN DISCOVERY
Data mining
Data mining uses computational and statistical methods to find patterns that may help answer a practical question. Useful work includes careful preparation, a clear target and checks that separate a repeatable signal from a coincidence.
Validate patterns on data not used to find them
Large datasets make it easy to find patterns by chance. Separate model development from evaluation, record the choices made during exploration, and check whether a pattern remains useful on new data. A pattern is not automatically causal, representative or actionable.
WORKED EXAMPLES
Data-mining guides
Methods are explained with examples and explicit limits.