Instance selection can be done by two methods, Sampling or by Search-based.
Examples for Sampling methods are:
 |
Random Sampling - randomly select "m" instances from the "n" initial instances. |
|
|
Stratified Sampling - randomly select "m" instances from the "n" initial instances, such that the distribution of classes is maintain in the selected sample. |
Examples for Search-based methods are:
|
|
Search for representative instances in the data, based on some criterion and remove the remaining instances. |
|
|
Form prototype instances from the actual instances, which would mimic the performance of these instances and then use only the prototype instances. |
|
|
Use Statistical measures (number of instances, mean or standard deviations) to replace redundant instances with their representative pseudo-instances. |
|
|
Use Support vectors to represent the entire set of instances from the data-set |