mentekid / metaclassifier

Implementation of an algorithm that combines the output of other classification algorithms to produce more accurate predictions

Metaclassifier

This repository implements the metaclassification process as described in the Metaclassifier paper

Install

The code itself requires no installation, you can run the code as-is.

The source code depends on these libraries:

• Numpy
• Pandas
• Scikit-learn, which you can install from here
• UnbalancedDataset which you can find on GitHub

Use

You can find usage samples in the Metaclassifier_Test*.py files. The Metaclassifier_Class has been implemented to look like a SciKit classifier, therefore any function accepting a scikit classifier as an argument can accept the metaclassifier as well.

Scikit's pipelines don't support a dataset with changing cardinality, therefore to implement SMOTE we had to reimplement the cross-validation procedure. This function is defined in GeneralCrossValidation.py