Skip to main content

Data reduction strategies

Short answer

Data reduction obtains a smaller dataset that yields similar analytical results. Strategies include dimensionality reduction (feature selection/extraction), numerosity reduction (parametric models, clustering, sampling), data compression, and data cube aggregation.

Explain data reduction with strategies.

Data reduction obtains a much smaller representation of the data set that produces the same (or almost the same) analytical results. It cuts storage and mining time on huge data.

Strategies

  1. Dimensionality reduction: reduce the number of attributes.
    • Feature selection (remove irrelevant/redundant attributes)
    • Feature extraction such as PCA and wavelet transform
  2. Numerosity reduction: replace the data by a smaller representation.
    • Parametric: fit a model (regression, log-linear model) and store only the parameters.
    • Non-parametric: histograms, clustering (store the cluster centres), sampling, data cube aggregation.
  3. Data compression: lossless (e.g. run-length, Huffman) or lossy (e.g. wavelets, JPEG) encoding of the data.
  4. Data cube aggregation: aggregate data to a higher level, e.g. quarterly sales instead of daily sales.
  5. Discretization and concept hierarchy generation: replace numeric values by intervals or higher-level concepts.
  6. Sampling: use a representative subset of tuples (see sampling methods).

Example

Daily sales for 5 years (1825 values) can be aggregated to 20 quarterly totals; a 100-attribute table can be reduced to its 10 principal components.

The reduction itself must be cheap compared with the mining time it saves.

✓ Answer checked by Claude Opus 5.5 (high effort)

Related questions

All Data Mining chapter 2 questions →