night18 / ChiMerge

Implement ChiMerge [1992] via python

ChiMerge

Implement ChiMerge [1992] via python

ChiMerge is announced by Randy Kerber in 1992. It is an algorithm that use X² statistics to discretize numeric attribution, which is a good method to reduce the afford of calculate before data mining.

ChiMerger consists 4 basic step:

1. Sort the data in ascending order.
2. Define initial intervals so that every value is in a separate interval.
3. Calculate the X² of any two adjacent intervals .
4. Find the smallest X², and merge the intervals who own the X²
5. Repeat 3 & 4 steps until all the X² is larger than threshold value.

The project use iris data as example, you can get the data from https://archive.ics.uci.edu/ml/machine-learning-databases/iris/