Finding useful patterns in a large database is a sequence of choices, not a single pass over every stored value. The records that matter are selected, cleaned, and transformed. Patterns are then mined from that prepared set, and the result is interpreted against the question that started the work.
The name for that full sequence is KDD. Data mining is one stage inside it. Treating the search as the whole process skips the work that decides which data is fit to search and whether a pattern deserves to be kept.
Usual Process Stages
The usual order is selection, cleaning, transformation, pattern mining, and interpretation. Each stage hands a narrower and clearer set of records to the next one.
| Stage | Work |
| Selection | Choose the records, fields, and time span that the question actually needs. |
| Cleaning | Repair missing values, duplicates, and inconsistent entries before any search. |
| Transformation | Reshape the cleaned records into a form the later search can use. |
| Pattern Mining | Look for relationships, groups, or repeated structures in the prepared data. |
| Interpretation | Decide which patterns are useful and state them in the terms of the original question. |
Data Mining as One Stage
Pattern mining runs only after the records have been chosen and prepared. It looks for relationships, groups, or repetitions. It does not decide which source to trust, and it does not decide that a finding is worth acting on.
- Bounded by selection. The search can use only the records and fields that were kept.
- Dependent on preparation. Cleaning and transformation change which structures are even visible.
- Separate from judgment. Calling a pattern useful happens after the search, not inside it.
Judging Whether a Pattern Is Useful
Interpretation puts the pattern back next to the original question. A result that cannot be stated with the source fields, or that disappears once obvious errors are removed, is not yet worth keeping.
- Matches the question. The pattern should answer the reason those records were selected.
- Survives basic repair. It should still show up after duplicates and inconsistent entries are handled.
- Can be explained plainly. Someone should be able to describe it with the original fields, not only with the output of the search.