COMB: A Hybrid Method for Cross-validated Feature Selection

When seeking to obtain insights from massive amounts of data, supervised classification problems require preprocessing to optimize computation. Among the various steps in preprocessing, feature selection (FS) empowers machine learning methods only to receive relevant data. We propose hybrid FS methods using unsupervised classification, statistical scoring, and a wrapper method. Among our tests using twelve dataset problems, the increase in performance from our novel method against existing FS methods represents an advancement in supervised classification.