Data Preprocessing

• Aggregation

  • Combine data objects to convert it into a higher resolution

• Sampling

  • Take a random subset of data from the entire population

  • A sample is representative if it represents the same statistics as the original data

  • Simple Random Sampling

    • Take out a random sample from the entire dataset

    • Sampling With Replacement

      • Put the taken object back into the population

    • Sampling Without Replacement

      • Take out the object and do not put back into the population

  • Stratified Sampling

    • Split the data into several partitions; then draw random samples from each partition

• Discretisation

  • AKA Binning
  • Process of putting continuous values into buckets so that there are a limited number of possible states

• Binarisation

  • Process of converting numerical or categorical data into binary values 0​ or 1​ based on a certain threshold

    ​​

• Attribute Transformation

  • Normalisation

    • Refers to various techniques to adjust to differences among attributes in terms of frequency of occurrence, mean, variance, range

  • Standardisation

    • Reshape the data to form a bell curve

      • Mean = 0
      • Std. = 1

• Dimensionality Reduction

  • Remove Irrelevant Features

  • Allow data to be more easily visualised

  • Avoid Curse of Dimensionality

    • When dimensionality increases, data becomes increasingly sparse in the space that it occupies

  • Techniques

    • PCA

      • Find a projection that represents the largest amount of variation in the data

        ​​

    • Feature Subset Selection

      • Remove Redundant Features
      • Remove Irrelevant Features

    • Mapping Data to New Space

      • Fourier Transformation
      • Wavelet Transformation