Current Slide

Small screen detected. You are viewing the mobile version of SlideWiki. If you wish to edit slides you will need to use a larger device.

Handling Training Examples with Missing Attribute Values

  • If data with missing attribute values exist, it is common to estimate the missing attribute value based on other examples for which the attribute has a known value.
  • Common strategy: Assign value that is most common among training examples at a certain node n that have the same classification as the current example.
  • More complex strategy used in C4.5: Assign a probability to each of the possible values.
    • Probabilities can be estimated based on the observed frequencies of various values of a certain attribute among examples at a certain node n, e.g., A(x) = 1: 0.6
    • Fractional of probability 0.6 is propagated down the branch of the tree for A=1.
    • Fractional examples are used to compute information gain.
    • Classification of a new instance is simply the most probable classification (computed by summing weights of instance fragments classified in different ways at leaf nodes of the tree).

Speaker notes:

Content Tools

Sources

There are currently no sources for this slide.