Extracting rules from DRASiW's "mental images"

DRASiW is an extension of the WiSARD weightless neural model that provides the ability of producing examples/prototypes, called “mental images”, from learnt categories. This work introduces a novel way of performing rule extraction by applying the WiSARD/DRASiW RAMbased neural model upon a well-known machine learning benchmark. A functional exploration is offered in order to demonstrate how the new rule extraction mechanism behaves under different system configurations. Experimental results suggest that the rules conformance to data increases proportionally to the corresponding classifier accuracy. Furthermore, comparison with C4.5 decision tree algorithm shows that the DRASiW-based technique produces more compact sets of rules.