Outlines

1. Implementation of paper, Fast Correlation Based Feature Selection (http://www.aaai.org/Papers/ICML/2003/ICML03-111.pdf)
2. Extract optimal feature sets using target, Symmetric Uncertainty

   Highly correlated to target
   Remove redundant features

from fcbf import fcbf
X = features
y = target
feature_set, history = fcbf(X, y, threshold=0, base=2, is_debug=True)
# `feature_set` refers tooptimal feature set
# `history` contains removed redundant features

Concepts

1. Optimal feature set consists of Predominant Features
2. Predominant Features are features that
   • (1) highly correlated to target
   • (2) there's no feature that correlated more than (1)

References

• Feature Selection for High-Dimensional Data: A Fast Correlation-Based Filter Solution (http://www.aaai.org/Papers/ICML/2003/ICML03-111.pdf)

TODO

• data to s3