Answer: KDD (Knowledge Discovery in Databases) is the complete, iterative process of turning raw data into useful knowledge: Data → Selection → Pre-processing → Transformation → Data Mining → Interpretation/Evaluation → Knowledge.
Diagram (in words): Draw boxes left to right joined by arrows:
Databases / Data warehouse → (Selection) → Target data → (Pre-processing / cleaning) → Pre-processed data → (Transformation) → Transformed data → (Data mining) → Patterns → (Interpretation / evaluation) → Knowledge.
Draw feedback arrows from the last stage back to earlier stages to show that the process is iterative.
Steps of the KDD process:
- Understanding the domain and goals: Study the application area, prior knowledge and what the user wants to discover.
- Data selection: Choose the relevant data sets, attributes and records from one or more sources (databases, files, warehouse).
- Data cleaning (pre-processing): Remove noise and errors, fill or drop missing values, handle outliers and remove duplicates.
- Data integration: Combine data from multiple sources into one consistent store, resolving naming and format conflicts.
- Data transformation / reduction: Convert data into a form suitable for mining, e.g. normalisation, aggregation, discretisation, and reduce dimensions or select features.
- Choosing the mining task and algorithm: Decide whether the goal is classification, clustering, association, regression etc., and pick a suitable algorithm and parameters.
- Data mining: Run the algorithm to extract patterns, such as rules, clusters, trees or trends.
- Pattern evaluation and interpretation: Keep only patterns that are valid, novel, useful and understandable, using interestingness measures; go back to earlier steps if results are poor.
- Knowledge presentation and use: Show the knowledge with reports, charts and visualisation, and use it in decision making or other systems.
Example: A supermarket selects its billing data, cleans wrong entries, converts it to customer-basket form, mines association rules such as "bread → butter", evaluates which rules are strong, and uses them to place products together.
Note: Data mining is only one step of KDD; KDD is the whole process from raw data to usable knowledge.